GE Energy Consulting

Final Report:

Technology Capabilities for Fast Frequency

Response

Prepared for: Australian Energy Market Operator

Prepared by: Nicholas Miller, project lead

Debra Lew
Richard Piwko
Contributions from:
Lou Hannett
Sebastian Achilles
Jason MacDowell
Matt Richwine
Douglas Wilson
Mark Adamiak

March 9, 2017

Imagination at Work
© 2017 General Electric International, Inc. All Rights Reserved.
Legal Notices
This report was prepared by General Electric International, Inc. as an account of work
sponsored by AEMO. Neither AEMO, General Electric International, Inc., nor any person acting
on their behalf:
1.! Makes any warranty or representation, expressed or implied, with respect to the use of
any information contained in this report, or that the use of any information, apparatus,
method, or process disclosed in the report may not infringe privately owned rights.
2.! Assumes any liabilities with respect to the use of or for damage resulting from the use
of any information, apparatus, method, or process disclosed in this report.

ii!
! !
!
Foreword
This paper was prepared by General Electric International, Inc. acting through its Energy
Consulting group based in Schenectady, NY, and submitted to AEMO. Questions and any
correspondence concerning this document should be referred to:

Nicholas W Miller
Senior Technical Director
GE Energy Consulting
1 River Road
Schenectady, NY 12345
+1-518-385-9865
Nicholas.miller@ge.com

AEMO Statement
“This report was commissioned by AEMO in June 2016 under its Future Power System Security
Program. The report forms part of the broad analysis AEMO is undertaking to identify and specify
potential technical solutions to current and future power system security challenges.”

iii!
! !
!
SYNOPSIS
This report discusses Fast Frequency Response (FFR) as a potential mitigation option for the
Australian Energy Market Operator (AEMO) in maintaining system security during low inertia
conditions on their system. The report examines (1) the practicality, requirements and
capabilities of several non-synchronous resources to provide FFR, (2) AEMO system
requirements and benefits of FFR, and (3) considerations for procuring FFR. The key study
findings are:

•! The following “unconventional” technologies may be suitable for providing FFR: wind
turbines; lithium, flow and lead-acid batteries; flywheel energy storage systems
(inverter-interfaced and non-inverter systems); supercapacitor energy storage
systems; solar photovoltaics (PV); load based resources; and high-voltage DC (HVDC)
transmission. Inertia-based FFR (IBFFR, also known as “synthetic inertia”) from wind
turbines can make a valuable contribution. PV could be an important FFR contributor
in future. Wind and PV can also contribute to primary frequency response. Both
synchronous and asynchronous flywheels have beneficial attributes that should be
considered.
•! The primary method to manage South Australia islanding should be by SPS (Special
Protection System). An SPS should also be designed to address related risks, including
loss of generation events (like the September 28, 2016 system black event). These low
inertia conditions chart new territory with respect to system dynamics and stability,
especially in South Australia. They should be addressed together with frequency
response.
•! Under low inertia conditions, identification of events must be quick and accurate. FFR
response time is critical, but a balance between making high fidelity decisions to act
and speed is needed. The analysis suggests that total response times on the order of
one quarter to one half second are sufficiently fast.
•! Detailed simulations are critical for determining the proper amount of FFR, and system
needs and operational practice for that FFR. As inertia decreases, it becomes more
important to provide the proper amount of FFR.
•! Load tripping or control will continue to be part of risk management for South Australia
and methods for this will need to evolve from present practice. Conventional under-
frequency load shedding (UFLS) should not be relied on for extremely high rate-of-
change-of-frequency (RoCoF).
•! It is important to ensure system security first, and base market constructs on that
foundation. Technologies with different energy delivery behaviors can be compared
for efficacy. Costs of FFR should be compared to the variable cost of alternative
operating strategies. Capital versus variable costs of FFR should be reflected in
procurement. Some services, like IBFFR, should probably be handled separately.
•! Achieving possible future conditions of zero synchronous generation, will require
significantly more infrastructure, and evolution of technology and understanding.

!!
! !
!
The key recommendations going forward are:

•! A comprehensive effort to specify and design a SPS to address separation of South

Australia from the rest of the National Electricity Market (NEM) should be a high priority.
The necessary system analysis and economic analysis should be performed to
establish the right mix of new FFR technologies and direct load tripping for the SPS.
•! Properly structured production simulations, that can capture these new constraints
and consider provision of FFR and FCAS from a full range of technologies should be
run. Economic evaluations, considering both capital and operating costs, are needed
to provide meaningful comparisons between options. FFR options should include the
technologies outlined in this report and conventional resources, as well as altered,
more conservative operating strategies and relaxed performance standards.
•! Model validation for evaluation of the dynamic performance of the NEM is needed, with
attention to whether the available models provide adequate fidelity for simulations of
the more extreme conditions under consideration. Performance of all important
elements, including renewable generation and all resources providing FCAS must be
considered.
•! Detailed dynamic analysis of the NEM is needed. Illustrative cases and relationships
developed with the very simple model of the South Australia system in this report
provide guidance for relationships to be more formally calculated.
•! For the near term, consider system solutions that will maintain a minimum level of
inertia and that will contribute other performance benefits as well (i.e. improved
transient, frequency and voltage stability).
•! The possibility for utility scale solar PV to provide low marginal cost FFR by taking
advantage of emerging practice of relative short-term ratings of panels and inverters
should be investigated (this is an industry-wide issue and recommendation).

i!
! !
!
 Table of Contents
1! EXECUTIVE!SUMMARY!...........................................................................................!1!
1.1! Project!Objectives!..............................................................................................!1!
1.2! Key!Findings!.......................................................................................................!2!
1.3! Summary!of!Recommendations!.....................................................................!13!
2! INTRODUCTION!TO!FFR!CAPABILITY!AND!TECHNOLOGY!..............................!15!
2.1! Fundamentals!of!Frequency!Response!.........................................................!15!
2.2! Basic!Components!of!Fast!Frequency!Response!.........................................!19!
2.2.1! Discussion!of!South!Australia!Situation!........................................................!20!
2.2.2! Response!Trigger!Options!..............................................................................!21!
2.3! Complexity,!Costs!and!Robustness!...............................................................!36!
2.4! Building!Confidence!in!What!is!Possible!.......................................................!37!
3! TECHNOLOGY!CAPABILITIES!FOR!FFR!SERVICE!............................................!38!
3.1! Overview!...........................................................................................................!38!
3.2! Wind!Turbines!..................................................................................................!39!
3.2.1! Overview!of!the!Physics!of!Inertia]based!FFR!from!Wind!Turbines!...........!39!
3.2.2! A!Note!on!Terminology!....................................................................................!40!
3.2.3! More!Discussion!of!Inertia]based!FFR!Controls!for!Wind!Turbines!...........!40!
3.2.4! How!Inertia]based!Response!Coordinates!with!other!Wind!Turbine!Controls
! 52!
3.2.5! Grid!Code!for!Inertial!Response!for!Wind!Turbines!.....................................!54!
3.2.6! Key! Technology! Considerations! from! the! Perspective! of! WTG!
Manufacturers!...............................................................................................................!55!
3.2.7! Synopsis!of!Wind!Turbine!FFR!.......................................................................!56!
3.3! Battery!Energy!Storage!...................................................................................!57!
3.3.1! Lithium!Batteries!..............................................................................................!59!
3.3.2! Flow!Batteries!...................................................................................................!64!
3.3.3! Lead]acid!Batteries!..........................................................................................!67!
3.4! Flywheel!Energy!Storage!.................................................................................!74!
3.4.1! Flywheel!Energy!Storage!Systems!(FESS)!with!Inverter!Interfaces!............!74!
3.4.2! Flywheel! Energy! Storage! Systems! with! Large! Synchronous! and! Near]
synchronous!Machines!...............................................................................................!76!
3.5! Supercapacitor!Energy!Storage!Systems!......................................................!79!
3.6! Solar!PV!.............................................................................................................!81!
ii!
! !
!
3.6.1! Solar!PV!Components!and!FFR!......................................................................!82!
3.6.2! Curtailment,!Overload!and!Provision!of!FFR!.................................................!85!
3.6.3! Inverter!and!Panel!Rating!................................................................................!90!
3.6.4! PV!Experience!with!Fast!Control!Response!..................................................!92!
3.6.5! Costs!.................................................................................................................!94!
3.6.6! Synopsis!of!Solar!PV!FFR!...............................................................................!94!
3.7! Load!Based!Resources!....................................................................................!95!
3.7.1! Inertia!Considerations!.....................................................................................!96!
3.7.2! Tripping!vs.!Blocking!vs.!Continuous!Control!..............................................!96!
3.7.3! Highly!Distributed!Responsive!Loads!............................................................!96!
3.7.4! Parallels!to!Industrial!Facilities!.......................................................................!97!
3.7.5! Synopsis!of!Load!Based!FFR!..........................................................................!98!
3.8! HVDC!Transmission!.........................................................................................!99!
3.8.1! Physics!..............................................................................................................!99!
3.8.2! Controls!...........................................................................................................!101!
3.8.3! Possibilities!and!Trade]Offs!with!HVDC!for!Fast!Frequency!Response!...!103!
4! IMPLEMENTATION!RISKS!AND!CONSIDERATIONS!........................................!107!
4.1! Torsional!Impacts!on!Turbine]Generators!...................................................!107!
4.2! Fault!induced!voltage!and!power!depression!.............................................!107!
4.3! Voltage!control!and!collapse!(weak!AC!system)!.........................................!109!
4.4! Differences!between!Under]Frequency!and!Over]Frequency!Response!.!109!
4.4.1! Risk!of!Conflicts!with!PFR!Droop!Response!...............................................!110!
4.5! Implications!and!Practicalities!of!Zero!Synchronous!Inertia!Operation!...!111!
4.5.1! WECC!and!other!Research!on!100%!Inverter]based!Systems!...................!111!
4.6! European!Rooftop!Solar!Experience!............................................................!112!
5! INTRODUCTION:! FFR! REQUIREMENTS! AND! ANCILLARY! SERVICE!
SUITABILITY!...............................................................................................................!114!
5.1! Arresting!Power!and!Energy!.........................................................................!114!
5.1.1! Frequency!Nadir:!the!metric!of!primary!frequency!response!....................!115!
6! ILLUSTRATION! OF! SOUTH! AUSTRALIA! FFR! REQUIREMENTS! AND!
ANCILLARY!SERVICE!SUITABILITY!.........................................................................!117!
6.1! Discussion!of!Model!and!Events!..................................................................!117!
6.1.1! South!Australia!Loadflow!Model!...................................................................!117!
6.1.2! South!Australia!Dynamic!Model!...................................................................!118!

iii!
! !
!
6.1.3! November!1,!2015!Event!................................................................................!118!
6.2! Requirements!for!South!Australia:!Present!Low!Inertia!System!...............!120!
6.2.1! Recent!History!for!South!Australia!...............................................................!120!
6.2.2! Context!using!example!of!November!1,!2015!event!....................................!122!
6.2.3! Sensitivity!to!Event!Size!(i.e.!Initial!Heywood!Loading)!.............................!128!
6.2.4! Other! Factors:! Short! Circuit! Strength! and! Dynamic! Performance! of!
Synchronous!Generation!...........................................................................................!130!
6.2.5! How!fast!is!fast!enough?!...............................................................................!132!
6.3! Requirements!for!South!Australia:!Future!extremely!low!inertia!system!.!133!
6.3.1! Stability!Risks!.................................................................................................!133!
6.3.2! Dynamics!of!extremely!low!inertia!system!..................................................!135!
6.3.3! Inertial]based!FFR!from!Wind!Generation!...................................................!135!
6.3.4! Summary:!Trade]off!of!FFR!requirement!with!Inertia!.................................!140!
6.3.5! Determining!Economic!Benefit!of!FFR!.........................................................!142!
6.4! Wind!with!IBFFR!and!PFR!control!................................................................!143!
6.4.1! Context!to!Other!Systems!.............................................................................!144!
6.5! Strategy!for!Extremes!of!Low!Inertia!and!High!Import!...............................!145!
6.5.1! Special!Protection!Scheme!summary!..........................................................!146!
6.5.2! Role!of!other!FFR!technologies!....................................................................!147!
6.5.3! Requirements!on!other!FFR!Technologies!..................................................!147!
6.6! Autonomous!Schemes!and!Backswing!Risk!...............................................!148!
7! FFR!AND!SIR!ANCILLARY!SERVICE!SPECIFICATION!.....................................!150!
7.1! Power,!Energy!and!Timing!............................................................................!150!
7.2! Evaluation!of!FFR!Performance!....................................................................!151!
7.2.1! FFR!Response!Efficacy!Mapping!..................................................................!152!
7.2.2! Evaluation!of!FFR!with!a!Complex!Response!.............................................!154!
7.2.3! Confirmation!...................................................................................................!156!
7.2.4! Non]Linearities!and!Other!Complexities!......................................................!157!
7.2.5! Applicability:!How!High!is!the!RoCoF?!........................................................!157!
7.2.6! Conclusions!of!FFR!Comparison!Method!....................................................!157!
7.3! A!Possible!Market!Construct!........................................................................!158!
7.4! Procurement!Discussion!...............................................................................!159!
7.5! Plant!Connection!Mandates!..........................................................................!160!
8! RECOMMENDATIONS!FOR!FUTURE!ANALYTICAL!WORK!.............................!161!

iv!
! !
!
8.1! Recommendations!for!Detailed!Dynamic!Analysis!.....................................!161!
8.1.1! General!............................................................................................................!161!
8.1.2! Class!of!Analysis/Cases!................................................................................!161!
8.1.3! Recommendation!for!Modeling!for!Detailed!Dynamic!Analysis!................!162!
8.1.4! Recommendations!for!other!technical!analysis!and!simualtions!.............!162!
8.2! Recommendations!for!Detailed!Economic!Analysis!..................................!163!
8.2.1! General:!Production!Simulation!...................................................................!163!
8.3! Zero!Synchronous!Generation!Future!.........................................................!163!
9! REFERENCES!......................................................................................................!164!
9.1! RoCoF!and!Frequency!Response!References!............................................!164!
9.2! Frequency]Input!Controls!for!Damping!.......................................................!165!
9.3! HVDC!...............................................................................................................!165!
!
! !

v!
! !
!
 Table of Figures
Figure 1 FFR Requirement vs. Contingency Size (for different system inertia) .................................. 8!
Figure 2 FFR requirement vs. Inertia for given disturbance size (with RoCoF noted) ..................... 8!
Figure 3 FFR Requirement Nomogram for Production Simulations of South Australia .............. 10!
Figure 4 Frequency response to a loss of generation event. ................................................................. 16!
Figure 5 Time Elements of FFR .............................................................................................................................. 24!
Figure 6 Example of commercial PMU response time and error in detecting frequency and
RoCoF at the start of a 1 Hz/s frequency ramp test .................................................................................... 28!
Figure 7 Details of the RoCoF error showing fast settling of RoCoF ................................................... 29!
Figure 8 Fast RoCoF sampling window may lead to false triggering ................................................. 30!
Figure 9 Impact of RoCoF Measurement Period on Measured Value ................................................ 31!
Figure 10 Bus Frequencies for Large Disturbance in USA Western Grid .......................................... 33!
Figure 11 Geographic Distribution of Bus Frequency for 2008 Florida, US Event ........................ 33!
Figure 12 November 1, 2015 Event (Measurement and Simulation) with Simulated Fault ...... 34!
Figure 13 Detail - November 1, 2015 Event .................................................................................................... 35!
Figure 14 Functional Representation of a Closed-Loop Inertia-based FFR Control .................... 43!
Figure 15 Field demonstration of one inertia-based FFR response 23................................................ 45!
Figure 16 Wind Inertial response (simulations) ............................................................................................. 46!
Figure 17 Sequence of Aggregate Response of an inertia-based FFR25 ........................................... 47!
Figure 18 Open-loop FFR Control (based on draft IESO requirement)................................................ 49!
Figure 19 Field Measurements of inertia-based FFR from Quebec27 ................................................. 50!
Figure 20 Frequency response for a Type 1 WTG in a system with low inertia ............................. 51!
Figure 21 Speed - Torque Curve NREL .............................................................................................................. 52!
Figure 22 Performance of HELCO BESS using frequency regulation algorithm. ........................... 64!
Figure 23 The Ecoult UltraBattery (hybrid supercapacitor and a lead-acid battery) .................. 68!
Figure 24 BESS supporting a trip to island event. ........................................................................................ 69!
Figure 25 BESS enabling resynchronization of island to main grid. .................................................... 70!
Figure 26 Notrees providing fast responsive reserves in ERCOT on Nov 1, 2013. ........................ 71!
Figure 27 BESS arrests under-frequency deviation in ERCOT. ............................................................... 71!
Figure 28 BESS arrests extreme frequency deviation in ERCOT............................................................ 72!
Figure 29 Basic PV module ..................................................................................................................................... 82!

vi!
! !
!
Figure 30 Illustration of DC Power Impact of Tracking .............................................................................. 83!
Figure 31 Illustration of Time of Year Impact ................................................................................................. 84!
Figure 32 Maximum Power Point Tracking Illustration.............................................................................. 84!
Figure 33 Inverter Rating Concept ...................................................................................................................... 86!
Figure 34 PV Current Phasor for Normal Operation ................................................................................... 87!
Figure 35 Normal Operation at Steady-state Current Limits ................................................................. 88!
Figure 36 Active Power given priority within Steady-state Rating ....................................................... 89!
Figure 37 Operation with Short-time Overload of Inverter...................................................................... 90!
Figure 38 DC power (different trackers) vs. AC Inverter Rating ............................................................. 91!
Figure 39 FFR capability for reduced rating PV inverters ......................................................................... 92!
Figure 40 FFR Tests for 20MW PV Plant ............................................................................................................ 93!
Figure 41 Response of BassLink to DC-Side Fault .................................................................................... 101!
Figure 42 Unstable Grid Event Involving HVDC Frequency-Input Modulation Controller ...... 102!
Figure 43 Primary Frequency Response (PFR) droop characteristic. ............................................... 110!
Figure 44 Overspeed event due to loss of load of 5 GW in Europe resulting in disconnection of
10 GW of DER at 50.2 Hz ........................................................................................................................................ 112!
Figure 45 Time Elements of FFR ........................................................................................................................ 116!
Figure 46 Sketch of Illustrative System Model ............................................................................................ 118!
Figure 47 Loadflow for November 1, 2015 event cases ........................................................................ 119!
Figure 48 Measurement and Simulation of November 1, 2015 Event ............................................ 120!
Figure 49 Recent (2016) History of Inertia in South Australia .............................................................. 121!
Figure 50 Recent (2016) Imports on Heywood Interconnector .......................................................... 121!
Figure 51 Inertia and Imports Correlation (2016) ...................................................................................... 122!
Figure 52 Simulation of November 1 Event without UFLS .................................................................... 123!
Figure 53 Frequency Response for Varying Inertia .................................................................................. 124!
Figure 54 Inertia and Primary Frequency Response Relationships .................................................. 125!
Figure 55 Illustration of FFR Requirement and Inertia ............................................................................ 126!
Figure 56 FFR Requirement to Avoid UFLS ................................................................................................... 127!
Figure 57 FFR Power Sensitivity......................................................................................................................... 127!
Figure 58 50% increase in event size ............................................................................................................. 128!
Figure 59 FFR vs. Frequency Nadir for different event size (fixed PFR and H) ............................. 129!
Figure 60 Minimum FFR necessary to Avoid UFLS (for one inertia and PFR condition) ........... 130!
vii!
! !
!
Figure 61 Effect of Other Parameters ............................................................................................................. 131!
Figure 62 FFR Requirement with Modified Dynamic Model ................................................................. 131!
Figure 63 Efficacy vs Speed of FFR for Low Inertia .................................................................................. 132!
Figure 64 Speed vs Power for Minimum FFR (to avoid UFLS) .............................................................. 133!
Figure 65 Stability Problem at Extremely Low Inertia - First Example............................................. 134!
Figure 66 Stability Problem at Extremely Low Inertia - Second Example ...................................... 134!
Figure 67 FFR equal to event size ..................................................................................................................... 135!
Figure 68 Comparison of 2 IBFFR Controls .................................................................................................. 137!
Figure 69 Illustration of Sensitivity to Balance at Low Inertia ............................................................. 138!
Figure 70 IBFFR benefits - rebalancing dynamics .................................................................................... 139!
Figure 71 IBFFR at Limit of Low Inertia and High Heywood Loading .............................................. 140!
Figure 72 Relationship of Inertia to Min FFR and Event Size ................................................................ 141!
Figure 73 Production Cost Constraints for Import and South Australia Inertia .......................... 143!
Figure 74 Illustration of Wind providing both IBFFR and Primary Frequency Response ........ 144!
Figure 75 Backswing Risk Illustration ............................................................................................................. 149!
Figure 76 "Ideal" FFR perturbation test .......................................................................................................... 151!
Figure 77 Frequency Response for Trip of Heywood Interconnector .............................................. 152!
Figure 78 Nadir Improvement vs. Timing of Arresting Power Injection .......................................... 153!
Figure 79 FFR Efficacy Mapping ........................................................................................................................ 153!
Figure 80 FFR Power with inertia-based FFR from Wind Generation .............................................. 154!
Figure 81 Aggregate Wind IBFFR Arresting Energy.................................................................................. 155!
Figure 82 Expected Component Benefit of FFR Energy ......................................................................... 155!
Figure 83 Expected Total Benefit from IBFFR Energy .............................................................................. 156!
Figure 84 Confirmation of Impact of IBFFR .................................................................................................. 156!
!
! !

viii!
! !
!
List of Acronyms
AC Alternating Current
AEMO Australian Energy Market Operator
AGC Automatic Generation Control
AS Ancillary Services
BA Balancing Authority
BESS Battery Energy Storage System
BMS Battery Management System
BOP Balance of Plant
CIGRE International Council on Large Electric Systems
CRS Contingency Reserve Service
CWFT Continuous Wave Frequency Transducer
DC Direct current
DER Distributed Energy Resources
DFG Double-fed Generator
DOD Depth of Discharge
DOE Department of Energy (U.S.)
DSM Demand Side Management
EMT(P) ElectroMagnetic Transients (Program)
EPRI Electric Power Research Institute
ERCOT Electric Reliability Council of Texas
FCAS Frequency Control Ancillary Services
FESS Flywheel Energy Storage System
FFR Fast Frequency Response
FRM Frequency Response Measure
FRO Frequency Response Obligation
GE General Electric International, Inc.

ix!
! !
!
HELCO Hawaii Electric Light Company
HVDC High Voltage Direct Current
IEC International Electrotechnical Commission
IESO Independent Electricity System Operator (Ontario)
LCC Line commutated converter
MPPT Maximum Power Point Tracking (for photovoltaics)
MW Megawatt
MWH Megawatt hour
NEM National Electricity Market
NERC North American Electric Reliability Corporation
NYISO New York Independent System Operator
OEM Original Equipment Manufacturer
PCS Power Conversion System
PFR Primary Frequency Response
PLL Phase-locked loop
PMU Phasor Measurement Unit
PSH Pumped Storage Hydropower
PSS Power System Stabilizers
PWM Pulse Width Modulation
PV Photovoltaics
RAS Remedial Action Scheme
RFP Request for Proposal
RoCoF Rate of Change of Frequency
RPS Renewables Portfolio Standard
SA South Australia
SCADA System Control and Data Acquisition
SCE Southern California Edison
SCR Short Circuit Ratio
x!
! !
!
SIR Synchronous Inertial Response
SNSP System Non-Synchronous Penetration
SONI System Operator for Northern Ireland
SPS Special Protection Scheme
SRP Salt River Project
SRS Supplemental Reserve Service
SVC Static VAR Compensator
UFLS under-frequency load shedding
VRB Vanadium Redox Battery
VSC Voltage Source Converter
VSPSH Variable Speed Pumped Storage Hydropower
WAMS Wide Area Measurement System
WECC Western Electricity Coordinating Council
WTG Wind Turbine Generator
WWSIS Western Wind and Solar Integration Study

xi!
! !
!
1! EXECUTIVE SUMMARY
1.1! Project Objectives
This report discusses Fast Frequency Response (FFR) as a potential mitigation option for the
Australian Energy Market Operator (AEMO) in maintaining grid security during low inertia
conditions on their system. Broadly, FFR is rapid injection of power or relief of loading that
helps arrest the decline of system frequency during disturbances. (A more detailed discussion
is provided in Section 2.1) The first part of this report focuses on the practicality, requirements
and capabilities of various non-synchronous resources to provide FFR. In the second part, we
examine system requirements and benefits of FFR, and provide discussion of considerations
for procuring FFR services.
Primary Frequency Response (PFR) is used in this report to denote the relatively fast,
autonomous reaction of system resources to change in frequency. In most power systems,
the main contributor to PFR is the governor response of synchronous generation. In the
National Electricity Market (NEM), the 6-second Frequency Control Ancillary Service (FCAS)
product is the means by which PFR is formally provided. However, throughout the illustrations
and discussions, we have used a generalized representation of governor response1.
To demonstrate the interplay of energy, speed, inertia, FFR and PFR, we provide illustrative
simulations on an extremely simple model of the South Australian system under a variety of
system conditions. One key element to maintaining system security in a power system with
low inertia is to provide adequate arresting energy to stabilize frequency. This theme is carried
throughout this work.
Broadly stated, AEMO’s goal is to maintain a level of system security consistent with both the
specific requirements of the Australian grid code (the National Electricity Rules), and with
accepted industry practice. The primary focus of this work is aimed at technology options
that have the potential to minimize or avoid involuntary under-frequency load shedding
(UFLS), which starts as frequency decreases to 49 Hz. A further objective is to advance the
overall security and robustness of the system, of which avoiding ‘system black’ events is a
significant consideration. The scope of this report has the narrow constraint of only
considering technologies outside of “conventional” solutions (“conventional” solutions might
include addition of synchronous generation, new transmission lines, synchronous condensers
or imposing operating constraints).
In the situation of South Australia (SA), there are credible losses, such as one circuit of the
Heywood interconnector. Further, there are non-credible events, such as the loss of both
circuits of the Heywood interconnector, which can result in the loss of up to (or sometimes
exceeding) the maximum planned infeed of 650MW to South Australia, which needs to be

1!This!assumes!a!closed1loop!response!(i.e.!controlled!increase!in!power!for!a!drop!in!frequency)!that!is!representative!of!how!
equipment!suppliers!tune!power!plants.!For!example,!we!have!used!the!IEEE!standard!model!GGOV1!for!the!responsive!generation!
in!South!Australia.!!The!question!of!exactly!how!the!present!resources!in!the!NEM,!and!particularly!in!South!Australia,!perform!is!of!
great!interest,!but!is!outside!the!scope!of!this!work.!

1!
! !
!
addressed separately. This is an extremely disruptive event, which has more severe impacts
on the SA system as import levels increase. The event severity increases even further under
low inertia conditions. While UFLS is allowed for these non-credible events, complete system
black must be avoided.
The quantitative illustrations and many observations throughout this report are based on the
situation in SA. SA was used as a case study for this analysis because the rate of change of
frequency (RoCoF) has been identified as an immediate challenge in SA3. While created with
the situation in SA as a significant consideration, the discussion is intended to provide more
general insight, applicable to the entire NEM. Issues of economically maintaining good
frequency response, with consideration of all available technologies to do so, is of interest for
the entire country.

1.2! Key Findings

Several “unconventional” technologies can provide Fast Frequency Response
Section 3 of the report examines the capabilities of the following technologies to provide
arresting energy for FFR:

•! Wind turbines
•! Lithium, Flow, and Lead-acid batteries
•! Flywheel energy storage systems (inverter-interfaced and non-inverter systems)
•! Supercapacitor energy storage systems
•! Solar PV
•! Load based resources
•! HVDC transmission
All were found to have the potential to provide FFR, but their characteristics, opportunity costs,
economics, and practicality varied. Some of these, e.g. batteries, flywheels and
supercapacitors, would be new resource additions that would need to recover costs by
providing FFR and possibly other services. Some of these, e.g. wind and PV, are being installed
for other reasons but would need to be procured with FFR specifications which may have
impacts on capital costs. Some of these, e.g., PV, may have an opportunity cost if they are pre-
curtailed to provide FFR. Table 1 summarizes some key characteristics of the different
resources.
Because FFR characteristics (speed, shape, duration) vary, this study created a new analytical
method to compare the effectiveness of different types of FFR (see Section 7). With this new
methodology, both planning and procurement can consider a range of technologies without
being unduly prescriptive of performance characteristics. Discussion is provided throughout
the report about physical limitations and practical considerations of specifying and obtaining
different performance from these FFR technologies.

3!https://www.aemo.com.au/1/media/Files/Electricity/NEM/Security_and_Reliability/Reports/FPSS111Progress1Report1August1
2016.pdf!

2!
! !
!
The application and opportunity cost for most of these technologies is straightforward. A
notable exception is that inertia-based FFR from wind turbines (IBFFR) does not need to be
pre-curtailed (or charged) to provide this service (although the energy recovery after IBFFR is
deployed needs to be considered). Because the available arresting energy with IBFFR is quite
limited, and because provision of this service by wind turbines has low operational and
opportunity costs, it may be appropriate to handle this type of service separately from FFR
technologies that can be sustained for longer periods. The aeromechanics of wind turbines
makes the practical upper limit of IBFFR control for wind turbines about 10% of the power
production. So, at an example level of 1000MW of wind power production, about 100MW of
IBFFR should be achievable.
PV could also be an important FFR contributor. The possibility for utility-scale PV to provide
low marginal cost FFR by taking advantage of emerging practice of relative short-term ratings
of panels and inverters presents a potentially economic option which should be considered.

Both synchronous and asynchronous flywheels have beneficial attributes that should be
considered. Synchronous flywheels (high inertia synchronous condensers) are a technically
effective means of increasing inertia and support grid voltage and short circuit strength.
Many of these resources, including wind and PV, can also provide PFR (FCAS in the NEM). This
can benefit system robustness. Provision of PFR (even for wind turbines, because PFR is not
inertia-based) typically imposes a substantial operating cost on them (in the form of lost
energy delivery, green certificates or other metrics associated with production), that needs to
be addressed in the procurement process.
Low inertial conditions require fast responses
A critically important characteristic of these resources is response time. Lower levels of inertia
on the system will result in higher RoCoF, i.e., for a given contingency, lower inertia means
frequency will drop faster. In order to arrest the system before frequency reaches UFLS trip
points, fast response time is required. Total response time is comprised of the following:

•! Measure – Measure and identify frequency deviation and fast frequency decrease.
•! Identify – Identify occurrence of severe event that requires FFR.
•! Signal – Communicate action to be taken.
•! Activate – Actuate the resource.
•! Activate fully – Full response from resource.
A summary of response times for the technologies above is presented in Table 2.

3!
! !
!
 Table 1. Summary of key metrics for each technology
Wind Lithium Flow Advanced Flywheel Flywheel Super- Solar PV Load HVDC
Turbines Batteries Batteries lead-acid (inverter) (non- capacitor Tripping
Batteries inverter)
Once 0.5-5 s Few – 40 Few – 40 Few – 40 ms ≤4 ms As per 10-20 ms 100-200 ms Depends 50-500 ms
triggered, ms ms synchronous on load
devices
time for
full
activation
Short-run Small for Very small Very small Very small Very small Small Very small None, if Depends None;
marginal inertia- (standby (standby (standby (standby (standby (standby overload. on load Opportunity
based FFR; losses) losses) losses) losses) losses) losses) Opportunity cost if
cost for opportunity cost if curtailed.
FFR cost with curtailed.
curtailment

Maturity Commercial, Rapidly Relatively Based on Commercial, Based on Relatively Relatively Relatively Mature, for
but limited moving to immature mature but limited relatively immature; immature for mature; triggered
deployments: mature technologies deployments mature evolving controlled technologies response
First technologies; rapidly active power for functions
generation rapidly evolving applications widespread
offerings of designs tripping of
‘synthetic many small
inertia’ loads
emerging

Other IBFFR is Capital costs Energy rating Can be Contributes High power May depend .
energy dropping highly attractive for short circuit density; tend on relative
important limited rapidly scalable applications strength. to be rating of
considerat with low Transient specialized, panels and
ions cycling (i.e. stability small inverters
contingencies) behavior applications
different. now

4!
! !
!
 Table 2. Summary of response times for various non-synchronous resources
Measure Identify Signal Activate Activate
Fully
Options for external detection and signaling
Direct detection ≤2-3 cycles ~1 cycle
40-60ms 20ms
RoCoF detection ~2-3 cycles ~1 cycle
w/PMU 40-60ms 20ms
Local ≥5 cycles nil
RoCoF/Frequency 100ms
measurement
FFR options that require external detection and signaling
Wind turbine with ~2 cycles ~500ms
inertia-based FFR 40ms
Lithium batteries 10-20 ms
Flow batteries 10-20 ms
Lead-acid 40 ms
batteries
Flywheels ≤4 ms
(inverter)
Super capacitor 10-20 ms
Solar PV 100-200 ms
HVDC 50-500 ms
FFR options that detect, signal and actuate
Flywheels (non- instantaneous N/A
inverter)
FFR options that signal and actuate
Load 2-3 cycles Depends on load
40-60ms

The main challenge is to quickly and accurately identify a severe event that requires FFR.
Complicating this step is the fact that directly after an event, frequency varies spatially. So,

5!
! !
!
while one part of the grid may perceive a severe event, another part of the grid may not.
Additionally, triggering too much FFR may have adverse consequences, as this work
demonstrates. Table 3 summarizes the tradeoffs between different types of detection and
topology of FFR response.

Table 3. Tradeoffs in topology and detection mechanism

Triggered by local RoCoF Triggered by direct event detection
measurement
Few, large Pros: Cheaper topology Pros: Low risk of false triggering
resources

Cons: High risk of false triggering Cons: Only applies for specific events;
Moderate cost for communications
Many, small Pros: Reduces consequences of Pros: Low risk of false triggering
resources false triggering

Cons: Expensive communication

Cons: Expensive detection
Another difficulty is that very fast measurements may misinterpret transients, switching
operations or other actions that are not severe events as reason to trigger. Risk of false
triggering is mitigated by longer periods for measurement and identification, but this comes
at the expense of FFR activation time. So, FFR response time is critical, however a balance
between making high fidelity decisions to act and speed is needed Fortunately, it turns out
that FFR needs to be fast but not incredibly fast. Simulations in Section 7.2 show the efficacy
of injections of FFR after a contingency event with varying delays in response time. FFR needs
to be started well before UFLS or the occurrence of the frequency nadir4,. Analysis presented
suggests that total response times on the order of one quarter to one half second are
sufficiently fast. It is non-intuitive, but extremely fast FFR is less effective. If it is too fast, then
it interferes with and stifles full PFR response. Part of the planning process can include fine-
tuning the response time of FFR, thereby improving the efficacy of the FFR for critical
conditions.

Fast Frequency Response, Primary Frequency Response, and Inertia all interact
There is a delicate interplay between FFR, PFR and inertia. The primary function of FFR is to
arrest the frequency decline and “buy time” for PFR1 to act. The amount of FFR needed and its
efficacy is closely tied to the amount and quality of PFR available. For example, faster PFR will
reduce the amount of FFR required at any given level of inertia; however at very low levels of
inertia, conventional PFR (from synchronous generation) has limited ability to provide arresting

4!Nadir!is!the!minimum!frequency!during!the!swing!from!a!system!disturbance!(see!Section!2.1).!

6!
! !
!
energy fast enough. Similarly, withdrawal of FFR should avoid abrupt steps, and should be
coordinated with the PFR.

Determining system needs and practice for FFR requires detailed simulations. This report uses
power system modeling to examine the interplay of inertia, FFR and PFR in detail. As noted, for
this work, a dynamic model of the South Australia power system was developed, with input
from AEMO. The model was designed to be relatively simple, but tuned to give the same
response as measured in South Australia for one event5. The model was exercised to
investigate the impact of speed and energy injected on FFR efficacy.
The largest factor in determining response is the size of the event, i.e., the amount of power
infeed lost in the disturbance. Figure 1 shows the FFR requirement as a function of
contingency size for two different inertia levels. The efficacy of adding inertia versus adding
FFR is given by the difference between the red and blue curves. For example, at 450MW import
(as indicated by the blue arrow), it takes about 120MW more FFR to avoid UFLS when the
system inertia is 1000 MW-sec than it does when the system inertia is 9000 MW-sec. Under
these conditions, the FFR must act within about 250ms for the 1000 MW-sec condition,
whereas this result for the 9000 MW-sec is based on 500ms action.
The economics can also be estimated from this. The capital cost to meet the marginal
requirement for FFR is the difference in cost of (about) 330 MW of 500ms response FFR
compared to the cost of 450 MW of 250ms response FFR. A very rough estimate for 120MW
dedicated energy storage device(s) would be A$120M (This is a high estimate, since at least
some of the necessary FFR could come from relatively less expensive resources, like IBFFR
from wind generation). The cost to add inertia (in the form of synchronous flywheels) is on the
order of A$35/kw-sec, so 8000 MW-sec, would be roughly A$280M. Note that this exercise
assumes a degree of linearity that may be at odds with the exact details of the system
performance. While this is a simplistic model with rough estimates, it suggests that new
infrastructure to provide inertia may need additional value streams to recover costs.
Getting the right amount of FFR becomes more important as inertia decreases. The amount
of FFR that is needed will be dominated by (1) the amount of load shedding that is allowed for
any particular operating condition and (2) by the size of the event. The ability to correctly shed
the right amount of load based on local measurement of frequency and RoCoF declines with
inertia. Presently observed levels of inertia in South Australia make autonomous UFLS of
questionable robustness. In particular, the use of highly distributed loads to provide FFR is
likely to be subject to the constraints and uncertainties of rapidly measuring RoCoF and
frequency.

5!System!separation!on!November!1,!2015.!

7!
! !
!
 900

800

700
0.23s
(200ms,FFR)
600
FFR#(MW)

500
0.31s
400

4.8s
300

200 5.4s
1s

100
6.1s

0
0 200 400 600 800
Contingency#size#(MW)

Inertia: 1,000,MW.s 9,000,MW.s

Figure 1 FFR Requirement vs. Contingency Size (for different system inertia). Data labels
show the time to the nadir in seconds.
160

140
1.93%Hz/s

120

100 0.97%Hz/s
FFR#(MW)

80
0.64%Hz/s

60

0.32%Hz/s
40
232)MW)contingency)size
500ms)FFR
20

0
0 2,000 4,000 6,000 8,000 10,000 12,000 14,000 16,000 18,000 20,000
Inertia#(MW.s)

Figure 2 FFR requirement vs. Inertia for given disturbance size (with initial RoCoF noted)

8!
! !
!
Simulations show how much FFR is required to avoid UFLS (49 Hz) for various levels of inertia
in South Australia (see Figure 2). For this case, the AC import that is lost when the Heywood
interconnector is opened is 232MW (approximately the amount lost in the November 1, 2015
event). As the system inertia decreases from 9000 MW-s to 3000 MW-s, the amount of FFR
required increases by about 60MW.
Costs of FFR should be compared to the cost of alternative operational strategies
In order to determine the relative marginal economic efficacy of adding FFR compared to
imposing operating constraints in the form of constraints on South Australia inertia and
imports from Victoria, detailed comparative production simulations are required.
Unfortunately, imposing dynamic constraints on production simulations can be challenging.
The results of dynamic simulations need to be mapped to static boundary conditions that can
be modeled in a production simulation.
There is industry precedent for this practice. For example, (in North America) transient stability
derived path ratings are often mapped into (so-called) nomograms. In Figure 3, a set of
nomograms show how a production simulation could impose the interrelated constraint of
import limit on the Heywood interconnector with committed synchronous inertia. The system
is assumed to be constrained by “available response”. We introduce this concept here,
because from a variable cost perspective (i.e. the constraints of a production simulation)
willingness to shed load by UFLS (or other means) and the amount of FFR available are
equivalent and additive.
The different traces are measures the sum of the allowed load shedding plus the amount of
FFR provided. For clarity, consider the blue “no response” trace. If we suppose that no load
shedding is allowed and that no resources are available to provide FFR, then this line
represents the loading constraint on the AC interface. At minimum inertia (of 1000 MW-s),
zero power can be imported. As inertia increases (e.g. by commitment of synchronous
generation in South Australia), then import limit eases, up to about 165MW at 9000MW-s. Now
consider the red “200 MW” line. If 200 MW of FFR is available, then at minimum inertia, the AC
import limit is 200MW. At 9000MW-s, the import limit is 330MW. But, 200 MW of load shedding
would produce the same result, as would (say) 100MW of FFR and 100MW of load shedding.
This is a production costing constraint. Other measures, including those that might require
capital expenditures (e.g. reactive compensation, or improved controls), could be needed, but
they have no bearing on the variable cost of operation. Using pairs of simulations with this
type of constraint allows a meaningful calculation or comparison of what FFR (or UFLS) is
“worth”, from a variable cost of operations (and emissions) perspective.

9!
! !
!
 Figure 3 FFR Requirement Nomogram for Production Simulations of South Australia
A properly designed production simulation will, at each operating interval (e.g. hourly), select
the most economic combination of committed generation and import (i.e. committing more
synchronous generation to adds inertia, raising the import limit as much as 175MW). If the
threshold is raised to (say) 200 MW of load shedding allowed or 200 MW of FFR available, the
red trace will apply instead. A production simulation with this eased constraint will result in
lower overall variable operating cost (VOC) of energy (on an annual basis, not necessarily at
each hour); it may result in lower carbon emissions as well. The difference provides a
calibration on the economic efficacy of relieving the constraint. That is, the results will show
what could be spent (annualized capital cost) on FFR, or what would be saved by allowing
more load shedding. This approach would allow for the evaluation of added inertia as well;
i.e. if dedicated flywheels were added to increase inertia all the time, the constraint would shift
to the left, rather than up. A similar comparison of annual VOC would show the energy and
emissions value.
The primary method to manage South Australia islanding should be by Special Protection
Scheme
This report finds that the most effective solution for managing the loss of the Heywood
interconnector will be a load tripping Special Protection Scheme (SPS), that is continuously
updated (e.g. by SCADA), armed and designed to interrupt load and trigger other discrete FFR
resources. The scheme should include some highly responsive, closed-loop continuously
acting FFR in South Australia. These resources need to be available (i.e. on line and not in

10!
! !
!
limits6) during periods of risk. The amount armed by such an SPS should be an amount close
to equal to the size of the event. The amount required will decrease somewhat with increased
inertia (see nomogram in Figure 3). Considerable further technical and economic (and societal)
analysis is required to establish the right mix of new FFR technologies and direct load control
for the SPS.
It is worth noting that UFLS does not help prevent separation due to overload of the
interconnector (e.g. the September 28, 2016 event). A suitably designed SPS would help.
Maintaining stability is more than just matching load and generation
It is critical to ensure system stability is maintained, especially in South Australia, while
examining options to manage under-frequency. That is, the South Australia system must be
able to transition to an island without loss of synchronism within the South Australia system.
The reactive power requirements of the system must be met. Good reactive power supply
and control can play a critical role in preventing an island from forming in the first place. The
system must have positive damping. Present technology inverter-based resources like wind
and solar PV tend to have better behavior with respect to power system oscillations. But
system dynamics change, sometimes dramatically with altered flow patterns and
commitment of synchronous generation. Inverter based devices must have sufficient short
circuit strength to maintain control stability. These considerations are outside the scope of
this report.
Further, the stability issues associated with islanding ought not to be addressed separately.
For example, as recent events have shown, large upsets, such as loss of generation within
South Australia, must also have stable outcomes. The best solutions to the islanding challenge
should be designed to produce benefits for other stability risks (beyond islanding) as well.
This is new territory with respect to system dynamics. The scale of the events in South
Australia, i.e. the amount of power that can be lost when the interconnector is heavily loaded,
relative to the total load being served in the receiving system, is extreme relative to most bulk
power systems. This report attempts to distill a large body of state-of-the-art technology
developments but further evaluation, testing, and analysis is required to ensure system
security in low inertia systems. Details matter, and the model used in this study is not a
substitute for real simulations with high fidelity models.
Zero synchronous generation is a near future condition that needs to be considered.
Technology, understanding and operational practice for zero synchronous operation are
outside the narrow scope of this FFR investigation. South Australia should, for now, proceed
with the understanding that there should be zero expectation today that it can survive,
islanded from the rest of the NEM, without at least some synchronous equipment in operation.

6!To!get!the!desired!response!from!a!particular!resource!(say!a!battery!energy!storage!system!for!example),!it!must!be!energized!
and!synchronized!with!the!grid!(“on!line”)!and!it!must!have!the!capability!to!respond!(“not!in!limits”).!!So,!to!continue!the!example,!if!
the!BESS!were!running!flat!out,!say,!selling!power!into!the!grid!for!arbitrage!revenue,!it!would!not!have!capability!to!provide!FFR!too!
at!that!instant!of!time.!!This!simple,!but!important,!constraint!can!be!easily!overlooked.!!

11!
! !
!
While the state-of-the-art is rapidly evolving, this condition will require significant work in
terms of infrastructure, understanding and adaptation of operational practice.

Load must continue to be a resource

Load tripping or control will continue to be part of risk management for South Australia. For
this discussion, fast voluntary disconnection of load, i.e. load tripping, should be considered a
legitimate type of FFR. This type of load control is a technically (and often highly cost) effective
option. This is a hugely important distinction from the type of autonomous, localized and
involuntary load shedding (i.e. UFLS) that is to be minimized. The presumption, from a
procurement perspective, is that load participation in FFR would be voluntary and
compensated. However, from a security perspective, the procurement aspect is a separate
discussion from that of effectiveness, economy and robustness.

For separation (islanding) of the South Australia system from the rest of the NEM – that is trip
of the Heywood interconnector or similar events (e.g. high loss of generation within South
Australia), economic solutions to maintaining acceptable performance will likely require that
service to some loads in South Australia be interrupted.
Methods of load tripping and control will need to evolve from present practice. Two important
factors will drive the evolution of load tripping from present day practice of actuating utility
(e.g. distribution feeders) towards more sophisticated methods. First, with the continued
growth of highly distributed generation, mainly PV, opening feeders disconnects generation
as well as load. It is entirely possible that the system, especially South Australia, will see
periods of zero net load in future. Second, as observed above, the requirement to have ever
more precise, real-time information about the exact amount of load reduction that will result
from each action grows with dropping system inertia.
Conventional UFLS cannot be relied upon to manage extremely high RoCoF. Historically, South
Australia has depended on UFLS that acts at or below 49Hz as a critical line of defense to
“save” the system. “Save” in this sense means that the frequency swing should not go below
47Hz, and the system should not go black. For the low range of inertia considered in this work
(i.e. down to 1000MW-s), this almost certainly optimistic. UFLS probably cannot be relied upon
to have the desired outcome for the very high RoCoF levels that will accompany simultaneous
low inertia and large disturbances (e.g. high loading on the AC interconnector). Therefore,
resources addressing low inertia conditions should be mostly, if not all, triggered by direct
event detection, rather than local frequency or RoCoF detection. Nevertheless, UFLS can still
continue to provide an important contribution to system security. It is envisioned that
conventional UFLS will continue to be used, and will serve to provide important protective
functions.
To design a robust UFLS solution, a detailed examination of the tradeoffs discussed here is
needed. Further work needs to go into:

•! Understanding the impacts of false triggering;

•! Understanding how easily severe frequency events in South Australia can be
discerned from other disturbances in South Australia; and
12!
! !
!
 •! Understanding how local frequency varies across South Australia due to a severe
event.
FFR procurement should consider both capital and variable costs
It is first important to note that physics should drive FFR procurement: maintaining the security
of the NEM, especially the South Australian system, is becoming increasingly challenging. This
is emphatically a power system dynamics problem that requires good utility engineering
practice. Market constructs must be founded on this, and must have sufficient flexibility to
adapt to what will inevitably be a rapidly changing situation – both in terms of available
technology and details of the grid dynamics.
Provision of FFR from most resources that might be built for the specific purpose of providing
that service involves significant investment in physical equipment. For dedicated resources
like batteries, flywheels, etc., the capital investment is the dominant cost. The variable cost of
providing FFR is usually relatively low. Other resources, which might be able to provide FFR in
addition to their main function (e.g. IBFFR from Wind, or FFR from temporarily over-driven solar
PV), may be able to offer the service after a relatively modest capital investment. Some
resources, like batteries, might also incur opportunity costs if they are performing other
market functions, such as energy arbitrage7. Design of a market for procurement of FFR needs
to consider these realities.

Procurement by an annual (or periodic) clearing market (i.e. winning resources are obligated
to provide FFR service when called upon, in return for a fixed revenue stream, plus energy net)
can be effective. Tendered procurement, in which a call is issued for competitive supply of FFR,
can also work. Variable (e.g. Day-Ahead clearing) markets for FFR may face challenges for
price formation and stability.

1.3! Summary of Recommendations

•! A comprehensive effort to specify and design a Special Protection Scheme to address
separation of South Australia from the rest of the NEM should be a high priority.
•! Properly structured production simulations, that can capture these new constraints
and consider provision of FFR and FCAS from a full range of technologies should be
run. Economic evaluations, considering both capital and operating costs, are needed
to provide meaningful comparisons between options. FFR options should include the
technologies outlined in this report and conventional resources, as well as altered,
more conservative operating strategies and relaxed performance standards.
•! Model validation for evaluation of the dynamic performance of the NEM is needed, with
attention to whether the available models provide adequate fidelity for simulations of
the more extreme conditions under consideration. Performance of all important
elements, including renewable generation and all resources providing FCAS must be
considered.

7!Energy!storage!may!participate!in!energy!markets,!buying!at!low!prices!and!selling!at!high!prices.!!Reserving!power!rating!for!FFR!
could!result!in!lost!revenues.!

13!
! !
!
•! Detailed dynamic analysis of the NEM is needed. Illustrative cases and relationships
developed with the very simple model of the South Australia system in this report
provide guidance for relationships to be more formally calculated.
•! Consider system solutions that will maintain a minimum level of inertia and that will
contribute other performance benefits as well (i.e. improved transient, frequency and
voltage stability).
•! The possibility for utility scale solar PV to provide low marginal cost FFR by taking
advantage of emerging practice of relative short-term ratings of panels and inverters
should be investigated (this is an industry-wide issue and recommendation).

14!
! !
!
2! INTRODUCTION TO FFR CAPABILITY AND TECHNOLOGY
Power systems are designed to maintain security of the system in the event of a contingency,
such as the loss of a large generator. This loss is invisible to the end-users, who see no
interruption in service or quality. In general, each interconnected power system should have
enough arresting power so that UFLS does not occur for most credible events8. UFLS, in which
blocks of firm load9 are disconnected from the grid to keep the rest of the grid operational,
starts at 49 Hz in the NEM.
Different power systems use different terminologies and different services with various
characteristics to arrest and then stabilize and recover frequency. The basic concepts remain
the same.

2.1! Fundamentals of Frequency Response

Key elements of frequency response as presently viewed by most of the power industry, are
shown in Figure 4. When generation is lost, frequency across the interconnected power
system will decline. Inertia from the rotating mass of generators and induction motors that
are online will determine how fast frequency falls immediately following the event. (This is
called the initial RoCoF). The first few seconds, up to the time of a frequency minimum (the
frequency ‘nadir’) is called the arresting period, and is shown in blue in Figure 4. The inertia of
the system is not a control action, but rather a physical characteristic of the rotating mass of
whichever synchronously connected devices are online at the time of the disturbance.
Without further action, the imbalance between load and generation will result in continuing
decline of frequency. Thus, inertia alone is insufficient. Historically, arresting the decline of
frequency results from the deliberate control actions taken autonomously by elements of the
power system: that is mainly individual generators with active speed governors and room to
increase output (headroom). This PFR will normally stabilize frequency in the seconds to tens
of seconds time-frame. While the figure shows this response starting immediately at the time
of the disturbance, it is important to note that it normally takes a second or so before any
additional power is injected to the grid due to governor action. And that it normally takes
several seconds, up to tens of seconds, before typical turbine-generators fully response to the
frequency error. In the NEM, there are a range of frequency control ancillary services presently
procured, including a 6 second product and a 60 second product. For the purpose of this
discussion, we are primarily concerned about the behavior of the system from the initiating
event up to about 20 or 30 seconds after the event. Therefore, it is (our understanding that)
the 6 second product that provides most of the PFR. As the size of system disturbances
increases, and as the amount of inertia decreases, the amount and speed of PFR needed to

8!Interconnected!power!systems!are!not!designed!to!withstand!loss!of!every!generator,!but!rather!are!designed!to!withstand!loss!of!
large,!credible!contingencies.!For!the!NEM,!loss!of!one!circuit!(e.g.!1!of!the!Heywood!interconnector!circuits)!is!considered!credible.!!
Simultaneous!loss!of!2!circuits!(e.g.!both!circuits!of!the!Heywood!interconnector)!is!not!considered!credible!by!NEM!rules.!Similar!
rules!exist!elsewhereU!for!example,!in!the!US!Western!Interconnection,!this!“designXbasis!event”!is!the!loss!of!2!large!nuclear!units.!
9!It!is!noted!that!with!the!significant!increase!in!behindXthe!meter!distributed!energy!resources,!“firm”!load!may!now!include!
generation.!This!new!twist!on!UFLS!schemes!is!important!to!consider!for!those!regions!where!distributed!generation!may!be!
significant,!and!deliberate!disconnection!of!that!generation!may!have!systemic!effects!counter!to!the!intent.!

15!
! !
!
arrest the system frequency decline before reaching a frequency that results in involuntary
disconnection of customers, i.e. ‘involuntary UFLS’, increases. Following the frequency nadir is
the rebound period, shown in yellow in Figure 4. The PFR will allow the frequency to settle at a
new point based on the load/generation balance.

Figure 4 Frequency response to a loss of generation event10.

The growing need for speed of response has given rise to consideration of a new, distinct
service: FFR. FFR is similar to PFR but acts much faster, providing power during the arresting
phase, with the specific objective of providing arresting power before the frequency nadir. PFR

10!Adapted!for!50!Hz!system!from!J.!Eto,!et!al.!“Use!of!Frequency!Response!Metrics!to!Assess!the!Planning!and!Operating!
Requirements!for!Reliable!Integration!of!Variable!Renewable!Generation”,!LBNLX4142E,!Dec.!2010.!
http://www.ferc.gov/industries/electric/indusXact/reliability/frequencyresponsemetricsXreport.pdf!

16!
! !
!
and FFR both help to arrest frequency and interact with inertia to determine the frequency
nadir. FFR will also contribute to establishing the settling frequency 11, if the FFR is sustained
past the time of the nadir into the rebound period. In the most common definition, both PFR
and FFR are autonomous controls that act based on local conditions, that is, they respond to
quantities like local frequency or machine speed) that can be measured at, or very close to,
the equipment providing the service. An example of the interplay between PFR and FFR in
ERCOT’s proposed ancillary services is described in the text box.
Secondary frequency response will increase output during the recovery period, shown in
green in Figure 4 to return frequency to nominal. Secondary frequency response occurs in a
few minutes. Unlike PFR and FFR, secondary response is often a centralized control action,
with the central energy management system of the grid operator sending power instructions
to individual resources (mostly participating generators) in response to deviation of system
frequency and interconnector flows. Finally, tertiary reserves will be dispatched up to allow
frequency regulation units to be dispatched back down and be available for the next
disturbance. Tertiary reserves often have market objectives, i.e. are “economic redispatch”.
Frequency response services are also used to mitigate over-frequencies, which could occur if
a block of load is dropped. In this case, generators respond by decreasing output. This is
discussed further in Section 4.4. While this is an important response and many of the
technologies in this report can respond well to both under- and over-frequency events, the
bulk of this report is focused on under-frequency mitigation which is a harder problem to solve
and the greater concern of AEMO.

11!“Settling!frequency”!is!the!period!during!which!the!PFR!has!stabilized!the!frequency!from!an!event,!but!before!the!secondary!
response,!i.e.!REG!(or!60!second!FCAS)!begins!to!restore!frequency!to!nominal!50Hz.!!It!is!represented!by!the!flat!section!between!
20!and!60!seconds!in!the!figure.!

17!
! !
!
 Future Ancillary Services in ERCOT
ERCOT is in the process of redefining their ancillary services (AS) as the ERCOT system evolves12
(changing generation mix, some of the new resources have new challenges and some of them
have new capabilities for AS, new regulatory requirements). Their proposed new services include
Synchronous Inertial Response (SIR), Fast Frequency Response (FFR), Primary Frequency Response
(PFR), Up and Down Regulating Reserve (Reg), Contingency Reserve Service (CRS), and
Supplemental Reserve Service (SRS) as shown below.

SIR is an instantaneous response that acts to decrease RoCoF, enabling time for PFR to arrest
frequency. FFR is a fast-acting response: the time from frequency meeting a threshold to full
response from the resource is 0.5 seconds. This is sustained for 10 minutes or until ERCOT recalls
the resource, whichever is less. FFR must then be restored in 90 minutes and ready to be deployed
again. FFR would be a new service, although 1400 MW of current Responsive Reserves from load
resources meet the FFR requirements. PFR is used to arrest frequency decay and provides
continuous full proportional response to frequency deviations. ERCOT has studied the interaction
between FFR and PFR. An example of how FFR may reduce PFR requirements is shown below.

Security constrained economic dispatch automatically dispatches generation every 5 minutes to

balance the system. Within those dispatch intervals, Reg follows an AGC signal every 4 seconds to
provide fast balancing of the system. CRS ensures that within 15 minutes of a disturbance,
frequency can be restored. CRS needs to provide 95% of full response within 10 minutes after
receiving ERCOT signal. Within 90 min of frequency restoration, PFR and Reg are restored and
ready to deploy again. Finally, SRS is a longer term product used to compensate for forecast errors.

12 The!ERCOT!process!is!onXgoing.!!Changes!have!not!yet!been!approved.!http://www.ercot.com/committee/qmwg!

18!
! !
!
2.2! Basic Components of Fast Frequency Response
Initial RoCoF is the rate-of-change-of-frequency immediately after the system is unbalanced
by a disturbance like loss of generation or islanding. Initial RoCoF is related to synchronous
system inertia and the size of the contingency according to the following equation (assuming
load dampening to be zero):
'( ∆+
"#$#% = = ×(
') 2- /
where
∆+ = 01234#(45#6)16736584(:;4<#=))4
- = 08=)3?4163@)1A44 :; − =35#6'=
(/ = %@3CD36584A)4)ℎ34)1?34#(4'1=)D@FA6534 -2
'(
= "A)34#(4$ℎA6734#(4%@3CD36584(-2/ sec)
')
In the instant following the disturbance, no control actions take part. Notice that the equation
does not include any terms other than the size of the event and the inertia. Therefore, for
systems where high RoCoF can drive adverse behavior, such as tripping of generators and
loads, high system inertia is desirable. There are two ways to manage frequency response:
1.! Slow RoCoF down: Synchronous generators provide inertia that decreases RoCoF.
Governors on conventional generators act fairly slowly (on the time frame of seconds
to minutes) to increase output to arrest and recover frequency. A conventional
enhancement would be to increase grid inertia using synchronous machines (e.g.
synchronous condensers) to slow RoCoF to allow time for other responses to activate.
Synchronous machines do this autonomously and with no need for ‘detection’ or
‘communication’. This is how the power system is managed at present and therefore
AEMO has directed that this not be the focus of this report.
2.! Respond more quickly: Another option is to mitigate the effects of fast initial RoCoF by
rapidly injecting arresting energy into the grid. Note that arresting energy could be
provided by increasing generation or decreasing load. Both provide the same function
of bringing system generation and load into balance. The objective of this study is to
examine what kind of very fast frequency response can be achieved using advanced
features from new inverter-based technologies such as (but not exclusive to) wind,
solar, batteries, flywheels. This option requires measuring frequency and identifying
the problem, communicating to the resource that a response is needed, and activating
the resource. This approach is the focus of this report.

12!!

19!
! !
!
2.2.1! Discussion of South Australia Situation
The South Australia situation is somewhat unique and is significantly more severe than what
has been experienced in other regions of the world. South Australia demand ranges from 800
to 3400 MW. This is demand on the grid, and excludes the approximately 680 MW of
distributed behind-the-meter PV. The Heywood AC interconnector has a maximum transfer
limit of 650 MW for flows in or out of South Australia, and is the region’s only synchronous
interconnection with the rest of the NEM. At maximum demand, loss of Heywood means a
19% loss-of-infeed event (650 MW/3400 MW). At minimum demand, loss of 650 MW
constitutes an 81% loss event (650 MW/800 MW).

Table 4. Time window for arresting energy before UFLS triggered

RoCoF Time before UFLS triggered Cycles13 before UFLS
triggered
1 Hz/sec 1000 ms 50
2 Hz/sec 500 ms 25
3 Hz/sec 333 ms 16.7
4 Hz/sec 250 ms 12.5

Two different types of events have distinctly different performance objectives14:

(1)! If a single generating unit trips, frequency must be maintained within the containment
band (49.5Hz) and the South Australia system should not experience involuntary load
shedding.
(2)! Simultaneous trip of both Heywood circuits should not cause full system black. Load
shedding is allowed. Trip of one circuit, when they both are in service before the event,
cannot result in tripping of the other circuit.
At high levels of loading on the Heywood interconnector, the simultaneous loss of both
interconnector circuits is on the extreme edge of typical utility practice. This situation is quite
challenging even before inverter-based renewables are added to the system. The relative size
of these loss-of-generation events is more closely akin to that faced by large industrial
facilities with internal dedicated generation (microgrids).
Calculations by AEMO have shown that if the Heywood interconnector is loaded at 650 MW
and both lines trip, the RoCoF can exceed 4Hz/s. Because UFLS kicks in at 49Hz, this gives the
system less than 250 ms (see Table 2) to arrest frequency before UFLS starts. The ability of
conventional UFLS relays to successfully disconnect the right amount of load, fast enough to
avoid unacceptable results (i.e. frequency below 47Hz and/or system black) is, at best, not

13!In!50!Hz!system.!
14!We!use!“objectives”!here.!!It!is!outside!of!the!scope!of!this!work!to!either!interpret!the!regulations!or!establish!precise!obligations!
or!ability!and!jurisdiction!to!act!to!satisfy!objectives.!

20!
! !
!
assured. AEMO is therefore examining what options can arrest frequency in this time frame
to ensure the RoCoF is slowed sufficiently for the UFLS to operate successfully.
This is pushing the envelope of existing requirements in other jurisdictions. ERCOT’s FFR
requires 500 ms; EirGrid/SONI’s FFR proposed requirement is 2 seconds; UK National Grid’s
Enhanced Frequency Response requirement is 1 sec15. (See box in Section 3.3.) The time for
which these various FFR requirements must be sustained varies, but the intent of all of them
is that the response last at least until the frequency nadir. Some pilot projects are
demonstrating fast response times with various technologies and these will be discussed in
the subsequent sections.
The overall system solution may include a combination of several of these FFR inverter-based
technology solutions together with some conventional synchronous solutions, and may also
include accepting some additional risk. But because of the severity of the challenge, and
because AEMO foresees a significant amount of new inverter-based generation being
installed in the near future, low cost solutions may include provision of these types of system
services from some of these new inverter installations.
It is important to view this issue with a realistic perspective. The worst-case events for the
South Australia grid, i.e. those that occur at high loading on the Heywood interconnector, are
significantly more severe (with much faster RoCoF) than what other similar grids are facing.
The solutions to the events are beyond existing experience, and may be beyond the existing
state-of-the-art applications of new technologies. New and innovative approaches will be
required. For this discussion, fast disconnection of load, i.e. load tripping, should be considered
a legitimate type of FFR. The concepts described in this report will each have some capability
to contribute arresting energy that improves the frequency performance of the grid. The
ultimate challenge is to balance the quantity and cost of the FFR resources with the achievable
improvement in grid frequency response. The baseline against which such solutions are
measured is the cost of reducing imports on the Heywood interconnector and running enough
frequency responsive generation (which must be available) to avoid unacceptable
performance. Today, frequency responsive generation means existing synchronous
generation, but should include the possibility of PFR from wind and solar generation as well.
Later in this report, some additional context for economic evaluation of operating costs and
benefits is provided, including participation by wind and solar generation. Involuntary load
shedding for extreme but rare events may continue to be part of a practical solution, provided
there are provisions for rapid restoration of load that was disconnected.

2.2.2! Response Trigger Options

There are two ways to detect the need to deploy FFR:
1.! Direct event detection – Detect the specific condition of the disruption (e.g. the relay
action that results in losing the AC link in South Australia) and having a direct transfer
trip scheme inform the resource(s). This can be done quickly (on the order of a few

15!National!Grid,!Enhanced0Frequency0Response,0Frequently0Asked0Questions0Version05.0,!29!Mar!2016.!!

21!
! !
!
 cycles) but it requires dedicated, fast communications and it only addresses the
specific contingencies within its design criteria (i.e., if something else causes a
frequency event, the FFR won’t trigger).
2.! Frequency and RoCoF detection – Detect the frequency deviation and high RoCoF.
There are promising new technologies that claim to be able to do this very quickly.
Accuracy and, especially, false triggering may still be an issue when attempting to
measure frequency and RoCoF very quickly after a major system fault, however, as
discussed below. Further, this approach has the limitation of only being applicable to
frequency events (and not, for example, excess interconnector loading that might
cause an island to form or other problems to evolve).
There are common elements to determining the potential for very fast response times. Figure
5 shows that the response time is the total of these actions:

1.! Measure (TMeasure) – The controller must first measure and identify the frequency
deviation and high RoCoF. It is important to note that local frequency may differ from
system frequency directly after a disturbance and implications of this are discussed
below.
2.! Identify (TIdentify) – Identify the occurrence of an event that requires FFR. This means
that the scheme must distinguish between a frequency event (for which action is
needed) and other events, such as a fault that successfully clears, for which FFR is not
needed.
3.! Signal (TSignal) – The controller must then communicate what action should be taken.
Communications latencies are too slow to allow for most multi-unit plant-level controls
to provide FFR. Instead, FFR controls may need to be implemented at the individual
unit-level in order to respond quickly enough. Algorithms to control the response will
also need to be fine-tuned to be very fast. This makes the use of distributed energy
resources (such as rooftop PV or small loads) difficult because detection and controls
will need to be distributed (in each individual PV inverter) and that cost may be
prohibitive. However, there may be ‘nearly instantaneous’ methods of
communications, such as powerline carrier approaches, that may reduce signal
latency times to workable levels.
4.! Activate (TActivate) – The controller then actuates the technology. This portion of FFR
performance is the time it takes for the technology to start acting (i.e., initiate its
response).
5.! Activate fully (TActivateFully) – The final measurement of FFR performance is not solely
related to time: The objective of FFR is to arrest the drop in frequency. This is
accomplished by delivering energy to the system. Thus, an appropriate measure of
performance is based on how much energy is delivered in the required time. Here we
are departing from some other proposals that focus on the rate of rise of the power
injection. We note the three different subregions of response in the figure, with
different hatching, and provide additional context for this approach in the subsequent
parts of the report. Response times of the technologies vary as described in Section 3.
A method for looking at the required response time and energy of dissimilar resources
is presented in Section 7.

22!
! !
!
Total response time, TTotalResponse, is then defined as:
KLMNOPQRSTMUSR = KVROSWXR + KZ[RUN\]^ + K_\`UOP + KabN\cONR + KabN\cONRdWPP^

Steps 1-3 are discussed below. Response times of various technologies (steps 4-5) are
discussed in Section 3. There we present technological capabilities and performance.
Performance testing has been documented for some projects, but it is noted that response
times can vary. For example, depending on when in the control loop cycle the event
occurs, the response could be signaled faster or slower.
In Figure 5, a simple illustration of the relationship between a frequency event and an FFR
response is shown. At the beginning of the event (1 second), system frequency begins to
drop at a RoCoF (as noted above) that is proportional to the size of the event, and inversely
proportional to the system inertia. As the frequency drops the event must be detected,
actions requested from FFR resources, they must respond by providing arresting energy
to slow and stop the frequency decline. A primary objective is to ensure RoCoF is
sufficiently reduced such that UFLS can operate successfully , so energy delivered before
hitting UFLS (green shading) is most valuable. Even if UFLS occurs, FFR continues to arrest
until the frequency nadir. This is the energy shaded in orange. After the frequency nadir,
the FFR energy complements the restorative energy coming from the primary frequency
response. This is shown in blue.
As the figure suggests, speed is important to avoid UFLS. However, we note that the power
industry has not asked for response times that are as fast as what AEMO is discussing.
Because of that, we note that technologies may not have been optimized for speeds less
than several hundreds of milliseconds. However, there may be modifications or trade-offs
that could reduce response times significantly. For example, an aggressive response from
lithium batteries may be at the expense of increased thermal stress on the battery cells
which could be mitigated by increased parasitic losses from increased cooling capacity or
sacrificing lithium battery life expectancy. Rotating devices like flywheels and wind
turbines will see higher torques, etc.

23!
! !
!
 Figure 5 Time Elements of FFR

2.2.2.1! Direct Event Detection

Direct Event Detection with Communicated Response Trigger

If the worst case contingency events are known, and if the type of response required to
mitigate the event is also known, then it is possible to directly trigger FFR that rapidly provides
arresting energy. These types of schemes are sometimes called Remedial Action Schemes
(RAS) or SPS. The basic approach involves:
1.! Detect the initiating event (e.g., breaker opening on the AC interconnector, unexpected
zeroing of current on the interconnector, etc.)
2.! Transmit that information to the FFR device (e.g., HVDC converter terminal or battery
storage system)
3.! Initiate a pre-programmed action to inject arresting energy into the grid.
This approach bypasses many of the challenges associated with rapid and secure
measurement of frequency and/or RoCoF. However, it introduces its own set of challenges,
which may or may not be easier to solve. Solutions are application-specific, and depend
primarily on grid topologies and communication system performance.

24!
! !
!
The topology of the problem for South Australia lends itself well to this class of solution. The
most severe grid event is well known (non-credible loss of the Heywood AC interconnector).
So, the major challenges with this approach involve communications. However, the fact that
there are a multiplicity of places where the system can separate means that the design will
be relatively complex. Further, as will be discussed at length below, the amount armed for
direct actuation must be continuously updated (e.g. by SCADA) for the current operating
condition.
The communicated signal is binary, so the message is very short. But the speed requirements
often require a dedicated channel from transmitter to receiver. Given the criticality of this
function to grid stability, redundant systems would be prudent, and are probably required to
meet reliability standards (as is the case in the US). State of the art communications between
major transmission substations are typically capable of such functions – as a custom
application within the communication network. But given the rapid speed required by AEMO,
actual performance capabilities of the South Australia communication systems would need
to be evaluated to determine feasibility. Single point-to-point high-speed channels are usually
straightforward to implement. Broadcasting signals to several substations may also be
possible with similar speed. Issues of cybersecurity must be addressed, and will likely involve
encryption.
There are operating installations, using commercially available technology, that appear to
meet much of the requirements for the South Australian challenge. For example, there is an
operating SPS in California for the Diablo Canyon nuclear power plant16. That SPS is designed
to detect when two geographically separated circuits both trip, and rapidly initiate tripping of
one of the 1GW Diablo Canyon units (so that the other unit will not lose synchronism). Total
time for detection and communication is less than 40 ms following the current zero.

Direct Event Detection with Wide-Area Broadcast Options

This concept is similar to the direct detection scheme described above, but rather than
transmitting the trigger signal to FFR resources located in major substations, this approach
would broadcast the trigger signal to many distributed resources spread over a wide
geographic area. The basic approach involves:
1.! Detect the initiating event (e.g., breaker opening on the AC tie; and/or current zero)
2.! Transmit that information to major transmission and/or distribution substations using
the system’s high-speed communication capabilities.
3.! Broadcast the message from the transmission substations to numerous individual
resources (e.g., demand side management (DSM)-enabled loads) using radio or other
technology
4.! Initiate a pre-armed action that contributes arresting energy into the grid (e.g., rapid
load-shedding).

16!“HighXspeed!control!scheme!to!prevent!instability!of!a!large!multiXunit!power!plant”,!Vahid!Madani,!E.!Taylor,!D.!Erwin,!A,!Meklin,!
Mark!Adamiak,!IEEE!2007!10.1109/CPRE!2007.359906.!

25!
! !
!
The major challenges for this type of scheme involve speed and security. Given that the
triggering signal needs to travel long distances and through several different communication
media in series, it may be challenging to meet the speed requirements for this application.
Cyber-security is also a challenge, since signal verification schemes would likely be necessary
to prevent false actions, and such schemes may introduce additional delay times.
However, with the assumption of well-developed and properly specified fiber optic
communications, there are commercial installations in operation which again appear to have
the necessary attributes for widespread activation of distributed and heterogeneous FFR
resources. For example, there is an operating SPS in Arizona , US, that is sensitive to the loss
of 2 of 3 of the nuclear generators at the Palo Verde Nuclear Power Station17. (This event is
the design basis event for frequency response in the Western Interconnection18 of the US).
That SPS is owned by one of the utilities that owns a share of Palo Verde, Salt River Project
(SRP). Upon detection of the units’ trip, signals to trip load are distributed to many SRP
substations. The decision processors have processing time of 1 to 4 ms, detection time for the
unit trip is about 1 cycle, and signal latency for communication to all the SRP substations is 7
ms. So, the measure, identify and signal steps (to the substations) is about 2 cycles. That
system is redundant and meets current North American Electric Reliability Corporation (NERC)
cyber-security rules. These systems can use a variety of protocols (e.g. jungle MUX; MPLS -
multiprotocol label switching).
Other more complex (and more expensive) systems can handle a wider variety of processes.
There is a new system in operation in Southern California Edison (part of the US Western
Interconnection around Los Angeles) that has a “line out” scheme. It detects breaker-open
and current-zero in 1 cycle (It is impossible to guarantee that the current will pass through
zero, allowing for interruption, faster than 1 cycle). This system uses secure GOOSE multitask
routable protocols that are authenticated and encrypted. Again, these meet all NERC
cybersecurity requirements. Enabling hardware for that system comes from SISCO19. Much
of the enabling protocols and practice are in the process of being added to the next version
of IEC Standard 61850.

2.2.2.2! RoCoF Detection

The focus of Phase 1 of this project is to examine various FFR technologies to determine how
well they can mitigate risks associated with high RoCoF. This requires detection of conditions
that require action. This section discusses some of the options and challenges with RoCoF
detection for the type of severe frequency event potentially faced by South Australia. Here we
address in detail the Measure and Identify steps discussed in section 2.2.2.

17!“Implementation!and!operational!experience!of!a!wideXarea!special!protection!scheme!on!the!SRP!system”,!J.!Sykes,!Mark!
Adamiak!G.!Brunello,!2006!Power!Systems!Conference:!Advanced!Metering,!Protection,!Control,!Communication!and!Distributed!
Resources,!10.1109/PSAMP.2006.285385.!
18!In!this!context,!“Interconnection”!refers!to!the!synchronously!interconnected!power!system!(the!western!portion!of!the!US!and!
Canada).!
19!www.sisconet.com!!

26!
! !
!
Measuring Frequency and RoCoF
The immediately obvious solution to the detection of severe events that will require the
FFR is to measure the frequency and RoCoF at the device or resource that will respond.
Frequency measurement at a generation resource differs, depending on whether the
generator is synchronous or non-synchronous. All synchronous generation uses the
speed of the machine as a proxy for system frequency. That is the origin of the word
“governor” – it governs the speed of the machine. There is a close relationship between
the speed of a synchronous machine and system frequency, but the two quantities are
NOT interchangeable when it comes to controls. In the time frame before speed equals
frequency, the inertial response of synchronous machines dominates. In this period, the
electrical power out is not equal to the mechanical power in. This inertial behavior is
dictated by the physics (Park’s equations) of the machine. It is not controllable; nor can it
be adjusted. Non-synchronous generation (such as wind and PV) must measure frequency
by some other means. This opens the potential for superior frequency response, but it also
presents an additional (and new) technical detail than can lead to unintended
consequences for overly prescriptive rules.
Therefore, measurement of frequency from observable quantities, i.e. voltage or current,
is more of a deduction than a direct measurement. Assuming that measured voltages are
sinusoidal20 (at 50Hz ± Δf), frequency deviation from 50Hz is effectively the rate of change
of the angle of the measured phasor. This means that frequency is effectively the
derivative of an observed signal, and therefore, it is noisy. That also means that the
measurement cannot be made instantaneously, since some filtering is needed to reduce
noise. There are a variety of algorithms that have different trade-offs in accuracy and
speed, e.g., time between zero crossings, fast Fourier transforms, etc. Control schemes
that have substantial system security impact and which depend on rapid response to
changes in frequency can require measurement systems specifically built and tested for
the application.21 Complicating matters further is that RoCoF is the derivative of frequency
with respect to time, and therefore the measurement is the second derivative of the
phasor angle. Because this is a difficult and noisy signal, it requires a significant sample
interval to achieve good fidelity.
It is noted that examples discussed in Section 3 are often triggered by frequency, not
RoCoF. Part of the reason for this is that some grids are less concerned with RoCoF and
more concerned with frequency. In some grids, RoCoF relays on generators are used to
protect them against anti-islanding, and therefore RoCoF is an important metric. In other
grids, frequency is the metric used to determine the health of the system and the main
concern is avoiding UFLS, so triggering on a frequency threshold is more appropriate. As

20!This!assumption!good!enough!for!most!purposes!here,!but!the!sinusoids!are!distorted!and!imperfect!during!severe!events,!such!
as!before!a!fault!is!cleared.!
21!Carlini,!EMU!Bruno,!GU!et.al.!“Electrochemical!energy!Storage!systems!and!Ancillary!Services:!the!Italian!TSO’s!Experience”,!C4X
116!CIGRE!2016,!

27!
! !
!
discussed above, an accurate measure of frequency is easier than of RoCoF. However,
there are issues with false triggering that are common to both frequency and RoCoF and
those are discussed below. Use of RoCoF has potential advantages as well. It is a leading
indicator of a frequency disturbance, and can improve dynamic response (not unlike
introducing lead to closed-loop control systems).

Phasor Measurements (PMUs) and Wide Area Measurement Systems (WAMS)

PMUs can measure frequency and RoCoF very quickly. For example, Vizimax’s PMU can report
out 200 frames per second and claim a highly accurate RoCoF calculation. They claim fast
settling of RoCoF in the time frame of 1.2 to 3.25 cycles (24 to 65 ms)22. Figure 6 and Figure 7
show laboratory test results of their PMU performance. (x axis is time in ms; FE is frequency
error, RFE is RoCoF error) If this is as fast and accurate when deployed on South Australia’s
system, then this may be a good solution, in conjunction with other fast technologies
described later in the report, for arresting frequency. Other original equipment manufacturers
(OEMs) for PMUs (e.g. GE) also report ability to measure RoCoF in 2 to 3 cycles. However, this
is still very much new ground for the industry, and consideration of the ability of these
approaches to differentiate local frequency disturbances versus systemic disturbances needs
custom evaluation for specific applications.

Figure 6 Example of commercial PMU response time and error in detecting frequency
and RoCoF at the start of a 1 Hz/s frequency ramp test23

22!https://www.vizimax.com/productsXservices/phasorXmeasurementXunit!!
23!Desrochers,!A.!“Vizimax!Energy!3.0!Phasor!Meausrement!Unit”,!March!15,!2016.!

28!
! !
!
 Figure 7 Details of the RoCoF error showing fast settling of RoCoF24
PMUs can measure system separation, rather than frequency. Work is being done in several
areas on this, including UK National Grid. The intent is to recognize when the system is
breaking into pieces, and to use that information to initiate controls that will respond properly
to the (sometimes) radically changed conditions that go with the island. This is essentially the
problem faced in South Australia. The PMU technology is well established, but the algorithmic
customization is still new ground. In these applications, the distributed PMUs require some
degree of communication between each other to make good decisions. This takes time.
Present objectives are (reportedly) to make these decisions on the order of ½ second.
Reaching decisions in less than ¼ second appears challenging, but progress is being made
rapidly on these fronts.

Identification Issues and False Triggering

There are two main challenges to fast RoCoF detection while minimizing false triggering:

1.! High RoCoF in one location may not be indicative of severe event – Assume that a
large FFR resource can quickly and accurately measure frequency and RoCoF at its
location and that based on this measurement, it identifies a severe event. However,
directly after a disturbance, frequency may vary locally. Depending on where this large
resource is located and how frequency varies across South Australia, it may
overestimate the severity of the event and over-respond or underestimate the event
and under-respond. A larger number of smaller FFR resources located across South
Australia may mitigate this issue but may be more complex and costly to implement.
2.! Distinguishing between severe frequency events and events that do not require FFR
– It is difficult to quickly distinguish between severe frequency events and events that
do not require FFR, such as a fault on the tie line that then clears. FFR that is fooled by

24!Desrochers,!A.!“Vizimax!Energy!3.0!Phasor!Measurement!Unit”,!March!15,!2016.!

29!
! !
!
 disturbances that are not severe under-frequency events may trigger falsely and
frequently, unduly impacting frequency. Allowing more time for the RoCoF detection
could mitigate this but that is at the expense of FFR response time.

Sampling Window Time

Another critical factor affecting measurement of RoCoF is the period of the measurement.
One issue is that a relatively long sampling window may be necessary to distinguish overall
grid frequency from local dynamic effects following a disturbance.
For example, Figure 8 shows a 20-degree phase jump event. This is a high, but credible
value of loss of one of the two Heywood circuits under heavy loading. Assuming frequency
and RoCoF are calculated at the minimum possible timing of every half cycle (10 ms), and
that frequency is calculated by measuring time between zero crossings, frequency of the
sampling window (red trace, right hand scale) after the event drops to 45 Hz. A 5 Hz drop
in 10 ms is -500 Hz/s RoCoF. RoCoF in the next sampling window would be + 500 Hz/s and
then zero after that. Triggering FFR based on a measurement of only the first sampling
window of 10ms would lead to an adverse response. Longer sampling windows would
increase the accuracy of identification of need for FFR. This is just one example of why
extremely fast measurement of RoCoF, even if the measurement is perfectly accurate, can
be problematic.

Figure 8 Fast RoCoF sampling window may lead to false triggering

30!
! !
!
 Source: EirGrid and SONI Position Paper, September 2012

Figure 9 Impact of RoCoF Measurement Period on Measured Value

See Figure 9, which appears in a paper prepared by Eirgrid and SONI25. In this example, a
sampling period of 500 ms gives a 1 Hz/sec RoCoF while a sampling period of 100 ms gives
a much higher value that is dominated by a local oscillatory phenomenon and is not
indicative of the overall grid frequency. AEMO will need to be cognizant of such effects
when attempting to determine whether to trigger FFR based on a rapid measurement of
frequency or RoCoF. A separate issue, which is the subject of a separate AEMO effort,
concerns the sensitivity of equipment to RoCoF. Unlike the deliberate response to
measured RoCoF, that is the subject of this investigation, if equipment (e.g. generation)
trips because of the high RoCoF in an undesired (and perhaps unexpected) fashion, this
high local RoCoF could presage systemic problems – including risk of cascading failure.
Frequency measuring devices that can meet the highly precise and fast requirements of
power electronic control during and immediately following system disturbances are often
specially created for the specific technology. For example, GE developed a Continuous
Wave Frequency Transducer (CWFT) for use in HVDC modulation functions and turbine-
generator power system stabilizers (PSS). This device uses phase-locked loop methods to
minimize sensitivity to non-fundamental frequency distortions in the input signal
waveforms. This transducer is directly integrated into these products to achieve fast and
stable performance in power swing damping applications. This type of transducer could
possibly be adapted for other types of applications.

25!!!RoCoF!Modification!Proposal–TSOs’!Recommendations,!!
http://www.soni.ltd.uk/media/documents/Archive/RoCoF%20Modification%20Proposal%20TSOs%20Opinion.pdf!!

31!
! !
!
If RoCoF measurement and event detection is used as a main method for triggering FFR,
it would be wise for AEMO to study their frequency response from real events to determine
the size of the sampling window to manage the tradeoff between accuracy in event
detection and time to respond.

Local Measurement of Frequency and RoCoF

Examples of events usually show system frequency to be a uniform quantity. However, in
the initial period following a large disturbance, system dynamics result in multi-modal
swings. Consequently, until these inter-area swings damp out, frequency varies with
location. This can be observed in Figure 10. The significance of this locational effect for
mandating and paying for frequency response, especially FFR, can be significant. Consider
the traces in Figure 10. The disturbance occurs at 1.0 seconds in the simulation. The buses
near the lost generation (teal blue) drop very rapidly (i.e. high initial RoCoF), but the bus
farthest from the disturbance, the orange trace here (which is more than 1000 miles away
from the disturbance), doesn’t “see” the event for nearly 3 seconds. A further example,
with time synchronized phasor measurements from a large loss-of-generation event on
the edge of the US Eastern Interconnection is shown in Figure 11. The farthest bus shown
does not see the event for about 4 seconds. It is of interest to note that the sign of the
frequency at this remote bus is the opposite of that on the buses nearest to the initiating
event after 3 or 4 seconds. Another excellent example of this effect for the Great Britain
grid is shown in a very recent CIGRE paper26.
These examples are for systems that are electrically and geographically larger than South
Australia. Consequently, the time for the event to propagate across these example
systems is longer than would be expected in South Australia, nevertheless, this information
demonstrates that FFR compliance based on when the event occurred, rather than when
the bus could observe the event, would have very different conclusions.
A discussed above, AEMO has performed analysis that shows a range of possible RoCoF
that is dependent on operating conditions, both present and possible in the future.
Measuring RoCoF and frequency with a view towards triggering controls for arresting
frequency collapse at the highest rates presently identified by AEMO as possible, i.e.
exceeding 4 Hz/sec, is largely uncharted territory. At the very least, detailed design and
evaluation studies would be required to qualify specific detection algorithms. There is
some precedence with bespoke technologies, such as those that have been developed for
individual HVDC projects and energy storage projects, but extending these concepts to
widespread deployment on a multiplicity of installations of different technology, vintage
and manufacturer, is, at present, unprecedented.

26!See!Figure!6!in!the!paper,!Advances!in!Wide!Area!Monitoring!and!Control!to!address!Emerging!Requirements!related!to!Inertia,!
Stability!and!Power!Transfer!in!the!GB!Power!System,!D.H.!Wilson!et!al,!CIGRE!Paper!C2X208,!Paris,!August!2016.!

32!
! !
!
 Figure 10 Bus Frequencies for Large Disturbance in USA Western Grid

Figure 11 Geographic Distribution of Bus Frequency for 2008 Florida, US Event27

27!Robert!Cummings,!NERC,!2012!“Florida!Disturbance!Feb!26,!2008”.!

33!
! !
!
Example of Challenge of RoCoF and Frequency Measurement Differentiation
Making extremely fast differentiation between critical events (for which FFR is required)
and other events can be challenging. Here a very simple example is presented. In Figure
12, three frequency traces are shown. The blue trace is a measurement provided by AEMO
of the November 1, 2015 separation event, and the red trace a simulation of the event,
performed on a greatly simplified model created specifically for this work. (The model
captures some of the essential elements of the South Australia system as it relates to the
separation and RoCoF issues, but is in no way intended to be a substitute for the detailed
representations maintained and used by AEMO for analysis. The model and more results
are presented and discussed at length in Section 6). This event is representative of the sort
of event for which FFR would be desirable. The RoCoF in this case is less than 1 Hz/sec.
The third trace is for simulation of a line fault and clear event, for which FFR would not be
necessary. Clearly, the green event represents a hugely different event.

Figure 12 November 1, 2015 Event (Measurement and Simulation) with Simulated Fault
But for this discussion, we are concerned with very rapid detection and response to the
event. So, in Figure 13 the same information is plotted, but zoomed in for the first 0.3s of
the event. Whatever was used to provide the AEMO frequency measurement (blue trace)
took about 80ms to respond, then it overshot, finally giving reasonable frequency
information after about 0.2s. This particular device could not have provided useful RoCoF
information faster than 1/5 second. The frequency traces are from the computer
simulation. The simulation uses positive sequence representation (as do essentially all
stability programs), and consequently frequency calculations are subject to the limitations
of sequence calculations and network solution methods. It is not obvious that the initial
jump up in frequency in the separation simulation is physically meaningful. The two
simulated events also have essentially identical frequency deviation (about -0.070Hz –

34!
! !
!
blue arrow) after 0.1s (red arrow). In any event, for this admittedly very simple example,
differentiation between these two dramatically different grid events in a time period less
than 100ms would be problematic at best. The point here is that, even with very high
fidelity calculation of frequency, dependence on purely local signals to rapidly detect
events may be subject to false triggers.

Figure 13 Detail - November 1, 2015 Event

Emerging Microgrid Technologies

There is an enormous amount of research and development under the umbrella of
“microgrids”. The technical challenges faced by South Australia have a high level of
commonality with the objectives and constraints of (at least some definitions of) microgrids,
and indeed one could argue that South Australia is (or becomes) a microgrid when it separates
from the rest of the NEM. Both the physical and MW scale of the problem is large (compared
to the subject of most work), but the issues of maintaining frequency and stability with a
predominance of inverter based resources and low inertia is essentially the same. The
proportional scale of the disturbances for which South Australia is trying to maintain stability
is nearly unprecedented however.

Virtual Synchronous Machines

The challenges introduced by displacement of synchronous generation by inverter-based
resources gives rise to a relatively simple question (with complex answers): “Why not make
the devices behave like synchronous machines?” And indeed, there is an emerging consensus
that inverter-based resources may need to emulate the electrical characteristics of
synchronous machines in systems that are predominantly or solely supplied by them. This is
not a new concept (see 1995 battery example in section 3.3), but has costs and complexities.

35!
! !
!
The issue is discussed further in section 4.5, but at present such controls are not (to our
knowledge) available in commercial wind and PV solar generation.

Other System Risks

The focus of this report is RoCoF for the NEM, with attention to the challenges associated with
separation of South Australia from the rest of the NEM. Given recent events28, it is worth
reinforcing that technologies introduced for managing high RoCoF should, ideally contribute
to providing other stability benefits, particularly helping to avoid separation of South Australia
from the rest of the NEM.

2.3! Complexity, Costs and Robustness

There are tradeoffs between the various approaches presented above. Cost, complexity,
speed, and risk of false triggering are all inter-related. Discussion above mentioned the direct
event detection versus local RoCoF detection methods. For example, detection that the AC
link is down is relatively inexpensive but costly communications links are required to transmit
that information to the FFR resources. In contrast, local RoCoF detection at the FFR resources
eliminates the cost of communication but at the expense of installing multiple RoCoF
detection devices.
Similarly, there is a tradeoff between the number of devices and the risk of false triggering.
Because local frequency may not be a good indication of a severe event, multiple RoCoF
detection devices across South Australia (associated with their own FFR resources or with
communications links to FFR resources) can reduce the risk of false triggering by giving a
better indication of system frequency health. This risk reduction comes at the expense of
higher cost in RoCoF detection devices or communications links.
And then finally, there is the risk of false triggering due to the difficulty of discerning an event
very quickly from the frequency measurements. The tradeoff in detection time and risk of
triggering when response is not needed must be considered. Because events that do not
require FFR can initially look like events that do require FFR, it may be better to give more time
for detection, as EirGrid has done with their proposed 500 ms averaging window. However,
this squeezes response time. For example, 500 ms would be too slow for AEMO to respond to
a 4 Hz/s RoCoF. There is not a specific “right” answer to this trade-off (e.g. “OK up to X Hz/sec
RoCoF”), as the figure implies. More quantitative context on the requirement for speed of
response is provided in Sections 5 and 6.
To design a robust solution, a detailed examination of the tradeoffs discussed here would be
wise. Further work needs to go into:
–! Understanding the impacts of false triggering
–! Understanding how easily severe frequency events in South Australia can be discerned
from other disturbances in South Australia

28!AEMO!“Update!report!–!Black!system!Event!in!South!Australia!on!28!September!2016”!!19!October!2016.!

36!
! !
!
 –! Understanding how local frequency varies across South Australia due to a severe event
–! Testing how well commercial solutions can detect severe events in South Australia
–! Quantifying the costs of various options

2.4! Building Confidence in What is Possible

In summary, AEMO’s needs for the South Australia grid for the most extreme events are at the
boundaries of existing technical capabilities and beyond industry experience. AEMO’s overall
speed of response targets are significantly shorter than those of others facing similar issues
(Great Britain, Ireland, etc.). Speed requirements on individual components that might be part
of an overall solution are also faster than previous applications, and may not be technically
possible with adequate resolution and security.

In the process of narrowing the possibilities and developing technically feasible conceptual
designs, laboratory testing of the essential components and subsystems is critical to success.
Performance characteristics of transducers, communication systems, and control systems
are application dependent, and cannot be derived from published specification sheets when
requiring performance at the limits of their capabilities. There is an axiom that seems
appropriate here; “One test is worth more than a thousand expert opinions.”

37!
! !
!
3! TECHNOLOGY CAPABILITIES FOR FFR SERVICE
This section describes technologies that may be able to provide a useful FFR and their
capabilities and characteristics.

3.1! Overview
There are many options for technologies that may be able to provide (the actuation portion
of) FFR. Because conventional synchronous technologies, (e.g. adding synchronous capacity
or adding new interconnectors) are well-understood, they are not included here. Instead, the
focus of this report is on inverter-based technologies and other technologies that may not
provide an inherent inertial response. The analysis is presented by technology in the following
sections. Ultimately, this is input to a broader evaluation of options for AEMO, that includes
the options identified in this report along with “conventional” alternatives.

Delivery of FFR services by the various technologies will depend on the physics of the
technology and on control. The distinction is important, and, is reflected in our discussion. In
so far as information is available, the capability of currently commercially available controls is
noted and described. However, it is important to recognize that provision of specific services
by technology is often catalyzed by market demands: even if no OEM has offered a
functionality, that capability might be offered if the market demanded it.
The discussion below also tries to provide insight into what is possible subject to limitations of
practical controls. Consideration of stability, latency, efficacy and robustness are included.
For context, it is of note that NERC has discouraged unduly fast response from inverter-based
resources in various instances due to the risk of unstable or interacting controls. “Just
because you can make it fast, doesn’t mean you should”. Good engineering practice and
reliance on good judgment is essential. It is challenging to make prescriptive rules, particularly
market rules, that will solicit and reinforce behavior by participants that best serves the needs
of a system. We have tried here to provide physical insight that will aid in making good
engineering decisions.

It is important to note that the problems facing South Australia are extreme. The phenomena
of interest are really fast. Insights as to what technologies will be successful won’t come solely
from equipment specification sheets. Further, equipment that can act rapidly under ideal grid
conditions, may not perform as well under weak grid conditions. The discussion provided
below, while created with the situation in South Australia as a significant consideration, is
intended to provide more general insight, applicable to the entire NEM. Issues of economically
maintaining good frequency response, with consideration of all available technologies to do
so, is of interest for the entire country.

38!
! !
!
3.2! Wind Turbines
3.2.1! Overview of the Physics of Inertia-based FFR from Wind Turbines

Most modern large-scale (>1 MW) wind turbine generators (WTGs) rely on sophisticated power
electronics to maximize wind power production, control turbine speeds and to provide a range
of performance functions. In these variable speed machines, the power electronic enabled
controls improve energy capture and provide characteristics that are beneficial to the grid.
The economics and physics of wind generation demand that the speed of rotation of the
blades must be adjustable with different wind speeds. The industry has almost unanimously
settled on the use of power electronics to allow the generator speed to vary, and this
electrically decouples the generator speed from the grid frequency. Consequently, unlike
simpler induction generator based systems, turbine-generators with these controls do not
naturally provide inertial response.

The existing population of WTGs in South Australia includes a mixture of different vintage and
technology wind turbines. Looking forward, essentially all new utility scale wind generation
being built in the world uses electronically enabled variable speed generation (i.e. Type 3 and
4).

In the case of induction machines and the truly synchronous machines (very rare “type 5” for
wind turbines), there is a direct connection between the power system and the machine. When
there is frequency decay on the power system, the induction machine will increase its output
temporarily because of the slip change. The induction machines make a limited contribution
to system inertia while the truly synchronous machines will inherently add inertia to the
system the same way a hydro or thermal turbine would29. Type 2 wind turbines, of which
there are some still in operation in the NEM, are induction machines with controlled resistance
on the rotor. This allows some degree of power control during swings. This is discussed
separately below in Section 3.2.3.3.

The basic design of the present class of converter based technology (Type 3 – aka “double-fed
– DFG” or “DFIG” generators) and 4 – aka “full converter generators”), however, does not
include any inertial response unless explicitly designed to do so. Both the DFG and full
converter generators employ a back-to-back converter to connect to the power system,
enabling speed variability for the turbine rotor.

Unlike conventional synchronous or induction generators, the delivery of active power from
variable speed wind generators is almost entirely controlled by the power electronics. The
power electronic controls allow nearly instantaneous adjustment of electrical torque on the
generator, and therefore power delivery, which is essentially independent of the terminal bus
voltage angle, and rate of change of angle – i.e. frequency. Controlled inertial response is
possible because of this extremely fast response. This is a fast, controlled response, rather

29!Kundar,!P.,!Power!System!Stability!and!Control,!1994,!McGrawXHill,!Inc.,!New!York,!ISBN!0X07X035958XX.!

39!
! !
!
than the inherent, uncontrollable response of all synchronous machines. This response is
transiently decoupled from the mechanical angle of the wind generator rotor. Therefore, it is
possible, though not necessarily desirable, to more-or-less instantly change (a few cycles)
electrical power delivery. As discussed above, in the context of the challenge facing South
Australia, this few cycles is the “activate” time. It is worth reiterating the note above (in
Section 2.2) that inverter-based devices, including wind turbines, could be designed to behave
as virtual synchronous machines. This presents a possible, even likely, path forward for the
industry faced with systems evolving towards occasional operation with zero synchronous
inertia. This is not the present offering, and this evolution comes with costs and challenges,
as discussed further in Section 4.5.

3.2.2! A Note on Terminology

The language of the industry is confusing and unsettled. Early work talked about “replacing”
the lost synchronous inertia from displaced thermal machines. Further, the available controls
generally are enabled by accessing the energy stored in the inertia of the turbine. So, in that
sense, these are inertial controls. BUT, the physics of the behavior is fundamentally different:
the power electronic devices DO NOT mimic Park’s equations…i.e. they don’t “look” like
synchronous machines. The power injected to the grid is one variety of FFR. Unlike some of
the other technologies discussed in subsequent sections, the energy available is relatively
limited, so the response cannot be sustained throughout the time period when secondary
frequency response returns system frequency to normal.

The discussion at times is becoming heated: some people at the US Federal Energy Regulatory
Commission strongly dislike the term “synthetic inertia”, but they also dislike simply calling it
FFR, since the inertially enabled control cannot be sustained. In contrast, what is generally
called PFR control from wind turbines is enabled by both the pitch control and the power
electronic torque control (discussed further below).

Through-out the balance of this discussion we will use the notation “inertia-based FFR” to
capture this specific behavior.

3.2.3! More Discussion of Inertia-based FFR Controls for Wind Turbines

The power delivery of the wind turbine-generator is limited not only by the available wind, but
by the physical limitations of the components of the WTG. Most critical are aero-mechanical
ratings and speed limits. The lift of wind turbine blades is a strong function of blade speed
relative to wind speed. These well understood relationships30 are reflected in the control
algorithms of modern, variable speed WTGs. Blade speed and rotational speed are directly
proportional, so turbine controllers target optimal rotational speed for power production. The
speed of the turbine is dictated by the mechanical torque delivered from the blades and the
electrical torque removed by the generator. When in balance, the turbine speed is constant.

30!Ackermann,!ThomasU!Wind!Power!in!Power!SystemsU!!WileyU!Sussex,!UK,!c.!2005U!pg!25X50.!

40!
! !
!
Inertia-based FFR Controls for wind turbines are based on temporarily making the electric
power delivered to the grid exceed the mechanical power being captured from the wind. The
source of energy for this extra power is the stored rotational energy in the turbine rotor and
drive-train. To extract that energy, the rotor must slow down.
A key point is that slowing the turbine tends to reduce the aerodynamic lift, thereby reducing
the delivered mechanical shaft torque and exacerbating the speed decline caused by
increased generator electrical torque. This positive feedback tends to push the blade towards
aerodynamic stall, which must be avoided. The inertial control must provide margin above
stall, and is consequently limited when the initial rotor speed is low. This means that the
energy available response is limited whenever the wind speed is at or below rated. Further,
the amount of energy available drops with wind speed (and therefore turbine power). At low
wind turbine power levels, the available energy of the inertial response starts to decline rapidly
below about 50% rated power, dropping to zero below about 20%.

Inertial energy extracted by slowing the rotation of the turbine must ultimately be recovered.
After the initial increase in electrical power, it must temporarily drop below the mechanical
power to allow the energy to recover, reaccelerating the rotor. The amount of energy that
needs to be recovered therefore includes (a) the extra arresting energy delivered to the grid
and (b) the energy NOT collected during the turbine speed drop because of reduced lift. One
can think of this second term as interest to be paid on the energy borrowed from the rotor.
This can be a critical consideration in systems with high penetration of wind PLUS limitations
on primary frequency response. Ultimately, inertia-based FFR from wind turbines buys time
for primary frequency response to act. This is a key fundamental observation: for operating
conditions at or below rated wind speed, the only NET value provided to the system from
inertia-based FFR essentially ends with the frequency nadir. That is, the value is in energy
delivered during the arresting period. Withdrawal of inertia-based FFR should, ideally, begin
at the nadir and be coordinated with the rate of rise of the primary frequency response during
the post-nadir recovery period. The coordination of withdrawal can be an important
consideration, if done too quickly relative to the rise in power injection from primary frequency
response, the frequency will fall again. But, done too slowly, the turbines will incur larger than
necessary rotational energy deficit that must be recovered.

Another aspect of the aero-mechanical rating of the wind turbine is maximum mechanical
loading. When the wind speed is above rated, the turbine controls reduce lift by decreasing
the angle of attack through pitch control. Under these conditions, the available wind power
is greater than turbine rated power. It is possible to increase the captured wind power, using
pitch control, to temporarily exceed the steady-state rating of the turbine. Thus, at higher
wind speeds, inertia-based FFR can transiently provide the extra power for inertial response
through pitch control - increasing wind power to the rotor, rather than just extracting stored
inertial energy by slowing the rotor. Under these conditions, the decline in rotor speed is less
and the energy recovery is minimal or non-existent.
The limitations of the electrical system must also be respected. The electrical rating of the
converter and generator are matched with the wind turbine. As with most electrical
machinery, the equipment has short-term capability that exceeds the continuous rating.
41!
! !
!
There are, nevertheless, current and voltage maxima that must be respected. The inertia-
based FFR control must respect these limits and coordinate with all other controls that can
drive the electrical loading on the converter and generator.
It can be a significant advantage that, unlike the inherent response of synchronous machines,
inertial WTG response is dependent on active controls and can be tailored, within limits, to the
needs of the power system. However, the response is shared with controlled variations in
active power necessary to manage the turbine speed and mechanical stresses. These stress
management controls take priority over inertial control. Turbulence may mask the response
for individual turbines at any instant in time, but overall plant response will be additive. It is
an important point for this discussion to note that in broad terms these constraints apply to
all modern variable speed wind turbines, regardless of manufacture. But since the primary
mechanism to deliver arresting power is via control, there is wide latitude available to the
equipment designers to implement different controls. In the following subsections, we will
discuss the two general classes of controls, which we have termed “closed-loop and open-
loop”.
3.2.3.1! Inertia-based FFR – Closed-loop Control Philosophy

Closed-loop inertia-based FFR features have been available commercially for many years31.
In this context, “closed-loop” means that the control action (i.e. active power injection to the
grid) is continuously varying based on continuous measurements (of frequency). Some
offerings were developed in response to the 2006 Hydro-Québec grid code32 to meet the
equivalent energy (kW-sec) contribution of a synchronous machine with 3.5 MW-sec/MW
inertia for the initial 10 seconds. Given the systemic needs, and the Hydro-Québec
requirement, the overall control that GE designed was intended to provide similar arresting
energy to that of a synchronous machine. A further objective of the control was to use the
finite energy available (i.e. as constrained by avoiding stall and respecting equipment limits)
to have the most beneficial impact on improving the frequency nadir. The behavior of the
control was therefore tailored to work well with a system that (a) has adequate, but relatively
slow, primary frequency response, and (b) reaches its frequency nadir in several seconds
following a loss-of-infeed event. The net result is a control that is rather different from the
inherent response of a synchronous machine.

Part of inertial control design constraints are to meet performance objective over most of the
turbine operating range. The specified inertia constant included in the Hydro-Québec grid
code is representative of large thermal generation. This target is met, for instance, when the
wind turbine dynamically varies the real power by about 5% for 10 seconds in response to a
large, short-duration frequency deviation on the power system. Thus, the energy delivered is

31!Nicholas!Miller,!Kara!Clark,!Reigh!Walling,!“WINDINERTIA™:!CONTROLLED!INERTIAL!RESPONSE!FROM!GE!WIND!
TURBINE!GENERATORS”,!MIPSYCON!09,!October!2009,!Minneapolis.!
32!“Transmission!provider!technical!requirements!for!the!connection!of!power!plants!to!the!Hydro!Quebec!Transmission!System”,!
March!2006.!

42!
! !
!
on the order of 0.3 MW-sec/MW when the turbine is operating near rated. Remember, this is
NOT the inertia (which has the same units), but the arresting energy that is delivered.

Hydro-Québec was the first transmission owner to require wind plants to contribute to
frequency regulation by using the inertial response. In 2010, Hydro-Québec started
integrating the first wind plants equipped with this feature in its network. Others in Canada
have followed, and several places in the US (notably ERCOT) are considering either
requirements or qualification of inertia-based FFR for inclusion in an FFR ancillary service
market.

The fact that some of these original designs were primarily based on the requirements for
Hydro-Québec has some relevance to Tasmania, as both systems have dynamics dominated
by the transient response of hydro generation. The performance discussion and examples in
the following section are based on a simplified representation of HQ (at low load and what at
the time was considered high wind).
A simple representation of the critical elements is shown in Figure 14. Frequency error is simply
the deviation from nominal. A positive frequency error means the frequency is low and extra
power is needed. The deadband suppresses response of the controller until the error exceeds
a threshold. This limits the inertia-based response to large events. The continuous small
perturbations in frequency that characterize normal grid operation are not passed through to
the controller. The frequency is extracted from the wind turbine phase-locked-loop (PLL). This
frequency and synchronizing signal is also used as input to the current control. The speed of
the frequency transducer relative to the speed of the control as originally developed, is very
fast – on the order of 50ms or less.
The Power Shaping block produces a driving power response signal. The power shaping block
includes a “lead” term; this is, in a sense, sensitive to RoCoF. This signal is further filtered and
coordinated with other WTG controls, particularly the generator power order in the Power
Coordination block. Finally, the inertial response signal is limited, sent to the turbine electrical
control which adds to the generator power order. This net command is implemented by the
WTG converter controls, ultimately resulting in power delivered to the WTG terminals.
Generator
Frequency Deadband Limiter Power
- Error Order
Terminal Power Power Turbine To
Frequency Shaping Coordination Controls Converter
+ WindINERTIA
Controls
Signal

Reference
Frequency

Figure 14 Functional Representation of a Closed-Loop Inertia-based FFR Control33

!

33!Nicholas!W.!Miller,!Kara!Clark,!Robert!W.!Delmerico,!Mark!E.!Cardinal,!“WindINERTIA™:!Controlled!Inertial!Response!from!GE!
Wind!Turbine!Generators”,!CanWEA,!Vancouver,!BC,!October!20,!2008.!

43!
! !
!
There are several differences between this inertia-based FFR response, and the inherent
inertial response of a synchronous machine. First, and most important, the control is
asymmetric: Inertia-based FFR (as offered by OEMs today) only responds to low frequencies.
High frequency controls are handled separately, by a different controller that can, if necessary,
provide sustained response, as discussed in Section 3.2.4.1. Second, the deadband ensures
that the controller only responds to large events – those for which inertial response is
important to maintain grid stability, and for which seriously disruptive consequences, like
UFLS, may result. It is important to note here that these two differences are not fundamental
physical requirements for inertia-based! FFR, but rather a result of the design objectives (i.e.,
help the grid for big loss of generation events).

Finally, an inertial-based FFR response means the speed of response is a function of the
control parameters. In the example shown, the response was tuned to provide good
coordination not only with inertial response of other generation on the system, but with
governor response of conventional generation as well. The ability to tune inertial response
(including shutting it off) provides the planning engineer with an additional tool to manage
system stability.

Field test results of this inertial control for various wind speeds on a single wind turbine are
shown in Figure! 15. These tests were run around 2008. The field data was generated by
repeated application of a frequency test signal to the control. The results, at various wind
speeds (as listed in the legend), were then averaged and plotted. These are tests of the IBFFR
only, so in each test, the turbines are producing the maximum power available for the wind
condition (i.e. they are not curtailed). Below rated wind speed (<14m/s) the results clearly
demonstrate the inertial response and recovery. Above rated wind speed the inertial response
is sustained by extracting additional power from the available wind (i.e. short-term overload
of the WTG). [These tests were on type-3 wind turbines, but the distinction between type-3
and type-4 is largely irrelevant for this function. The difference in implementation between
OEMs is more important.]

These cases all have an initial rate of power rise of about 3.3%/sec (meaning % of rating, not
actual power; this works out to 50kw/sec for this individual turbine). At the time this work was
done, the upper design limit for rate-of-rise of power was 20%/sec. So, for this (AEMO)
discussion, the gains could be increased to have that rise time. Other OEMs have claimed
possible rise times up to 30%/sec. Since inertial response for this design is limited to 10% of
nameplate, the turbine can reach full response in 0.5 seconds, once the control is activated
by crossing the frequency deadband.

44!
! !
!
 Power

2000

1500 5m/s
8m/s
kw

1000 10m/s
11.5m/s
500 14m/s

0
0 20 40 60

Time (seconds)
Figure 15 Field demonstration of one inertia-based FFR response 23

Similar results from Vestas are shown in simulations of Figure! 16. For the event, the figure
shows a range of responses, as is typical for wind turbines running with actual wind
conditions. The response of turbines is of the order of 7 or 8%, with about 5% of the rise
occurring over a period of less than one second. The variability in response is an important
reality of individual wind turbines. Dictating exact behavior of a plant, and certainly of
individual turbines, is not likely to be possible with any available wind turbine control
technology. It is also of note that the Vestas authors are dismissive of RoCoF sensitive controls,
regarding them as insufficiently robust: “…makes this type of control a non-viable candidate”.

45!
! !
!
 Figure 16 Wind Inertial response (simulations)34
The diversity of response is shown in the sequence of traces of Figure 17. The simulated traces
are from the control work by RePower (now Senvion) performed in conjunction with Hydro
Quebec.35 Note that all WTGs are subject to the energy recovery period, which for these
examples occurs after about 10 seconds of response. The various control schemes employ
slightly different schemes, but all authors point out the risks associated with excessive

34!Peter!W.!Christensen,!German!Tarnowski,!“Inertia!for!Wind!Power!Plants!–!State!of!the!Art!Review!2011”,!12th!Wind!Integration!
Workshop,!2011.!
35!M.!Dernbach,!D.!Bagusche,!S.!Schrader,!“Frequency!Control!in!Quebec!with!DFIG!Wind!Turbines”,!12th!Wind!Integration!
Workshop,!2011.!
!

46!
! !
!
recovery power. The following discussion provides additional context to this important
consideration.

Figure 17 Sequence of Aggregate Response of an inertia-based FFR35

3.2.3.2! Inertia-based FFR – Open-loop, triggered Control Philosophy
The need for extremely fast rates of response for South Australia may drive a requirement for
an approach that provides a predetermined response to a triggering signal. As discussed
above, the trigger could be based on local information such as measured RoCoF or frequency,
or from a communicated triggering signal from elsewhere. (The considerations outlined above
for measurement versus direct communication apply here). The key distinction is that the
47!
! !
!
response, or more specifically at least the initial response, is predetermined once the trigger
occurs. There is precedence for this type of control, since it is a form of SPS. In the case of
South Australia, making closed loop control fast enough for the trip of the AC interconnector
raises concerns over control stability and unintended consequences. Nevertheless, open loop
responses are not without drawbacks as well. There can be (institutionally) a tendency to
focus designs on worst case scenarios, which will drive very aggressive behavior. This can
result in unintended, and undesirable, outcomes for other normally less interesting events.
Risks of overshoot, unnecessary stress (including torsional duty not only on the turbine, but on
other rotating equipment on the grid) must all be considered. In any event, the analytical
burden of either option is not trivial.
One type of open-loop control is shown in Figure 18. This figure is the PFR requirement by
Ontario’s Independent Electricity System Operator (IESO), recently added in June, 201636. The
orange annotations were added by us. The figure represents two “cases” of response. The
idea is to apply a minimum performance requirement for wind plants. “Case 1” is for a
frequency event that is sustained for more than 10 seconds. The basic idea is that the WTG
increases its electric power by 10% over the power production level that immediately
preceded the event. The turbine power is required to ramp up to at least 110% of this Po within
1 second, and sustain it for 10 seconds before being allowed to ramp down below Po. In “Case
2”, the control is allowed to revert to the recovery phase when the frequency returns to within
a small deadband of nominal. The areas on the figure represent, respectively, the extra energy
delivered by control (Area 1) and the recovery energy (Area 2) not delivered to the grid to allow
the WTG to return to normal speed. The amount that power drops below Po in the recovery
phase is noted as !Pr. The IESO notation suggests that this be no more than 10%.
The “area 1” is almost, but not quite, the arresting energy delivered by the control. The
difference is that the “release” of the control is based on frequency returning to normal, and
not on the time of the frequency nadir (which coincides with RoCoF passing through zero and
becoming positive. The figure carries some serious risks for practical application. Specifically,
the relationships between “Area 1”, “Area 2” and “!Pr” are emphatically not independent. The
physics of the problem dictate that the longer the turbine runs below its optimum speed, the
more recovery energy is needed. Thus, as !Pr shrinks, Area 2 increases, and the time until the
turbine recovers (Tr) increases substantially. If !Pr is too small, the turbine will never recover
and will stall. This is true for all variable speed wind turbines. At least some manufacturers
(good data is scarce) have tended to make !Pr rather large, in order to minimize both Area 2
and Tr. But there is a penalty on the grid for erring in either direction: recovery that is too fast
(i.e. !Pr too large) will starve the grid of power too soon, but recovery that is too slow (i.e. !Pr
too small) will increase the uncaptured energy and exacerbate the system frequency recovery.
In the extreme case !Pr too small will place the turbine at risk of stalling.

36!IESO,!Market!Manual!2:!Market!AdministrationU!Part!2.20:!Performance!ValidationU!Issue!5.0.!

48!
! !
!
 Figure 18 Open-loop FFR Control (based on draft IESO requirement)
The exact behavior will vary somewhat between manufacturers, and with the level of Po. For
calibration, the absolute minimum !Pr that one OEM’s turbines can tolerate is on the order of
7-8% of Po.

The consequence of this complication is that, in our opinion, the inertia-based FFR should be
designed to start “withdrawing” at the frequency nadir, and not wait until the frequency has
actually recovered. Incurring the extra rotational energy deficit associated with continued
inertia-based FFR after the nadir does not produce a significant performance benefit, as long
as the release is not too abrupt. The open-loop control philosophy drives a very “bimodal”
response from these resources: with no response at all for events below the trigger threshold
(e.g. inside the deadband), and the same aggressive response for all events that meet the
triggering criteria. Such abrupt discontinuities in power system control have been known
cause problems.
The FFR control implemented by Enercon for Hydro Quebec37 is a triggered response, which is
released with some hysteresis, i.e. when the frequency recovers to a given level (0.05Hz) above
the trigger frequency, it begins to ramp back over a period of time (5 seconds). The triggered
release avoids unnecessarily sustained response. The results shown in Figure 19 are
illustrative (there are several more good traces in the paper).

37!Markus!Fischer,!Sonke!Engelken,!Nikolay!Mihov,!Angelo!Mendonca,!“Operational!Experiences!with!Inertial!Response!Provided!
by!TypeX4!Wind!Turbines”,!Wind!Integration!Workshop!13,!2014.!

49!
! !
!
 Figure 19 Field Measurements of inertia-based FFR from Quebec37
The closed-loop control addresses most of these worries, with a gradient of responses related
to the error (i.e. to the amount that event passes the threshold for triggering). This allows the
triggering threshold to be set closer to “normal” with less fear of adverse consequences from
unintended triggering. But as noted above, very high closed-loop gains may raise stability
worries. It is possible that a hybrid of open and closed loop controls is appropriate. For
example, an open-loop “kicker” on the closed-loop control can discretely “force” the control to
act aggressively in response to specific stimulus (e.g. a trigger from an SPS). This maintains
the benefits of a closed-loop control, while making better use of the equipment capability for
extreme events – like separation of South Australia from the NEM. There is precedence with
HVDC, and in the downward direct for wind plants. Upward kick is NOT the present state of
the art, although the industry in some places is evolving in that direction – i.e. very aggressive
open loop response when frequency thresholds are passed.
3.2.3.3! Frequency Response via Torque Control on Type 2 Wind Turbines
Type 2 wind turbines are no longer being built by Vestas, but there are installations on the
NEM. Like Type 1 machines, these machines are electrically coupled to the grid and are

50!
! !
!
sensitive to system frequency. Standard induction generator equations apply. Figure 20
shows the behavior of a type 1 WTG in a system with rapidly dropping frequency and low
inertia. The initial upward spike in power is the inertial response, and it is beneficial to the
system. The recovery swing is not, in general, helpful.

Figure 20 Frequency response for a Type 1 WTG in a system with low inertia38

In the case of a Type 2 machine, there is a degree of control gained by having fast control (via
a power electronic chopper circuit) over the external field resistance. In Figure 21, a family of
speed-torque curves is shown. The black trace represents the “normal” condition, with the
external resistance (total rotor resistance is shown in the legend) shorted (only internal rotor
resistance is left). The machine will see an increase in slip from the pre-disturbance level
(suggested by the blue dot) as the grid frequency falls, increasing torque (as suggested by the
blue arrow). Increased toque will cause more power to be delivered to the grid, a benefit
similar to that from synchronous inertia. If the initial operating point were on one of the
colored curves, the torque could be further increased, added to the arresting power provided
by the WTG. This is essentially an FFR service, but it comes at the expense of having more
rotor resistance, therefore more loses, and higher rotor speed. Higher rotor speed means
faster blade tip speed, which may (or may not) reduce the power being captured from the
wind. Therefore, there is the possibility of a non-trivial efficiency penalty associated with
providing this control. Specifics of the turbine and generator would need to be considered.
There is also the possibility of using the field resistor to ease the power recovery after the
frequency nadir. Once the WTG has slowed, as the grid frequency recovers post nadir, the

38!Muljadi,!E,!et!al!–!NREL!“Understanding!Inertial!and!Frequency!Response!of!Wind!Power!Plants”,!July!2012!NREL/CPX5500X
55335.!

51!
! !
!
torque will drop allowing the rotor to recover the energy lost to the grid. This has similarities
to the recovery of the type 3 and type 4 WTGs. The field resistor control will allow some control
of the motor toque during this phase (as suggested by the red arrow), such that the recovery
power (see the backswing in Figure 20) is less acute and the time of the recovery is extended.
Like the other machines, there are aero-mechanical constraints that must be respected. It is
not known whether this type of control has been commercially offered for type 2 WTGs.

Rotor Speed (RPM)

Figure 21 Speed - Torque Curve NREL39

3.2.4! How Inertia-based Response Coordinates with other Wind Turbine Controls
Unlike the inherent response of synchronous machines, inertia-based WTG response is
dependent on active controls. The primary control actuators are generator torque and pitch
control. These actuators are also dedicated to meeting other control objectives for the WTG.
Variations in wind speed, and particularly turbulence, constantly change the torque and other
mechanical forces on the drive train, blades and entire structure of the turbine. The same
actuators that provide inertia-based response are also relied upon to manage the turbine
speed and mechanical stresses. These stress management controls take priority over inertial
control. As noted above, rating of the turbine constituents must be respected. During more
violent events, such as transmission system faults, functionality that provides fault ride-
through capability to wind turbines will take precedence of FFR function. Measuring frequency

39!Bastos,!A.F.,!et!al!“Use!of!Newton’s!Method!for!RotorXResistance!Control!of!Wind!Turbine!Generators”!!ICREPQ!’12,!Spain.!

52!
! !
!
(much less responding to it with a specific active power response) during a severe fault is
technically untenable for all technologies.

Further, the transient dynamic range of the actuators is dependent on the operating condition
of the wind turbine. For example, at lower wind speeds the generator and converter will tend
to have additional ‘room’ to increase power, BUT the speed of the turbine may be lower,
meaning that there is less inertial energy stored and less speed margin to aerodynamic stall.
Conversely, at wind speeds above rated, there is additional power available from the wind, so
that extra energy (power) can be captured from the wind, rather than slowing the turbine
drivetrain. However, the time dependent power ratings of all the components, especially the
converter, must be respected, and these considerations apply to all OEMs, regardless of their
approach to implementing Inertia-based FFR.
3.2.4.1! How WTG inertial response coordinates with pitch enabled Primary Frequency
Controls
The active power controls offered by some wind OEMs typically work through the wind power
plant’s supervisory control system. The controls typically impose curtailment orders, ramp
rate limitations, and PFR40. PFR is the autonomous frequency sensitive controls closely akin
to governor controls for thermal and hydro generation. These controls also respond to
significant deviations in grid frequency at the wind plant, increasing (“raise”) or decreasing
(“lower”) power output in response to low or high grid frequency events, respectively. Slower
controls that respond to external signals (usually from the grid operator), like Regulation or
contingency dispatch, also work through the plant supervisory system. Unlike inertia-based
FFR controls, these controls alter the active power control reference targeted by the turbine
controls. Therefore, they can, under some circumstances, have significant impact on energy
production. (Further discussion of primary frequency control is provided below in Section 4.4.)
In the NEM, there are a range of FCAS’s. These services are related to PFR, but have their own
characteristics. In so far as these services provide active power response in proportion to
frequency error, they are equivalent to PFR. While FCAS functions can clearly be provided by
this type of control, it is not a given that IBFFR will seamlessly integrate with them. This is part
of the necessary detailed analysis recommended.
When enabled, the response of the active power control to significant under-frequency events
is to increase WTG output. Usually, the command for this response emanates from the wind
plant level control, and is delivered to each individual WTG. To increase active power, the
plant must be partially curtailed, so that additional power can be extracted from the available
wind. This incremental power order signal will add to the power from inertia-based FFR
control which is local to the individual WTG. The total response of the WTG to these two signals
is coordinated to respect the physical capabilities of the WTG. Again, whether these controls
reside in the plant level controller or in the individual turbines, depends on the manufacturer.

40!Cardinal,!M.EU!!N.W.!Miller,!“Grid!Friendly!Wind!Plant!Controls:!WindCONTROL!–!Field!Test!Results”U!!proceedings!WindPower!
2006,!Pittsburgh,!PA!.!

53!
! !
!
And, as noted above, approaches to these controls vary considerably by OEM, and not all
vendors even offer this class of control.

It is important to reinforce that, in general, these active power controls reside at the plant
supervisory level. This means that signals from the host utility (e.g. curtailment orders) as well
as systemic measurements are processed at plant controller, and individual signals (usually
power set points) are sent to the individual turbines in a wind plant. This has many
performance and security benefits (e.g. the plant controller understands when individual
turbines are unavailable, and can compensate), but it requires communication from the plant
controller to the turbines. This takes time. Communications latencies vary with plant size
(count of turbines), OEM and vintage. A reasonable range of communication latencies for
present technology commercial wind plants is on the order of 200 to 500ms.
Grid over-frequency events present a different risk to system operations. Excessive high
frequency is stressful to power components. Further, temporary high frequency swings can
present a reliability concern. For example, in one recent well publicized grid event41, the high
frequency backswing from a major grid disturbance caused power plant trips and aggravated
an already severe event. When enabled, the primary response (at the plant level control) will
reduce power output for the duration of the over-frequency event. This behavior is similar to
that of governor control on thermal generation, except that it is faster and allows deeper
runback of power than is typical of thermal generation. Having this function has been
mandated in EirGrid and ERCOT (Texas) for years. A broader discussion of backswing over-
frequency is provided below, in Section 6.6.

3.2.5! Grid Code for Inertial Response for Wind Turbines

Ultimately, system operators have begun modifying grid codes to include some type of inertial
response requirement. The development and demonstrations by several OEMs, shows that
such functionality is, indeed, possible. However, it also shows that inertial response identical
to that of synchronous generation is neither possible nor necessary. Controlled inertial
response of wind plants is in some ways better than the inherent inertial response of
conventional generators. Inertial response of wind generation is limited to large under-
frequency events that represent security and continuity-of-service risks to the grid. The
crafting of new grid codes should therefore proceed cautiously and focus on functional,
systemic needs.
The following lists some elements that should be included (from our perspective) in sensible
FFR technical requirements for wind turbines:

•! Specification of minimum turbine power below which inertial response is not required
•! Specification of preferred interactions/priority with other wind plant controls (E.g.,
should FFR or curtailment have priority? Should reactive support or frequency support
have priority?)

41!FRCC!Event!Analysis!Team!(FEAT)!Interim!Recommendations!ReportU!!!!
http://www.balch.com/files/upload/FRCC_Interim_Report_6_3_08.pdf!!

54!
! !
!
 •! Not specified to be identical to synchronous machine
•! Not specified to be identical for all operating conditions
•! Not specified to be exactly reproducible with individual turbine tests
•! Not specified to be energy neutral (for all events)
•! Not specified to be overly prescriptive or requiring of impossible power recovery
constraints
•! Not specified to require delivery of energy beyond that necessary to improve the
frequency nadir

3.2.6! Key Technology Considerations from the Perspective of WTG Manufacturers

This section is intended to provide some insight into how wind turbine manufacturers (OEMs)
will view new performance requirements. It is, of a necessity, somewhat speculative and
based on our experience.

3.2.6.1! Limits to Performance

What is it that OEMs will worry about when asked for very fast and aggressive inertia-based
response? They will concentrate significantly on turbine speed and mechanical loading. In
general pitch response isn’t the limit; under the conditions when the turbines are pressed to
meet performance objectives, the blades tend to be at pitch limits anyway. Even for combined
performance (with governor/PFR behavior), the blades will tend to catch up. Likewise, the
inverter itself is very fast. The limitation on response is much more related to interaction with
other controls in the turbines. For example, very aggressive response stimulates the drive
train dynamics. Active loads management in (all/most) modern WTGs will tend to respond to
abrupt changes in torque, and these controls are likely to want to defeat the inertial-based
FFR controls. Likewise, turbulence and loads management are considerations that take
priority over grid functions. Therefore, demands that a certain level of response be provided
under ALL conditions will inevitably be met with conservative response from OEMs.
This raises several interesting questions about requiring, providing and paying for
performance. Basic reality is that if resources are rewarded for doing the best they can at
any instant, the result will be better. It presents an interesting riddle: is this (“do the best you
can, and you will be rewarded”) an appropriate response to “non-credible” events (like loss of
both circuits of the Heywood interconnector)? Is this a legitimate response to design-basis
(credible) events? Or should AEMO handle the two classes of events differently? This is part
of a bigger industry debate.
OEMs will worry a great deal about the frequency with which the turbines are subjected to
stressful control actions. They will ask: How many times/year would the control trigger? For
example, an OEM is likely to get nervous at about 100/year. Repeated actuation starts to
have fatigue costs; these are very difficult to quantify, but will affect willingness to have
aggressive response.

55!
! !
!
3.2.6.2! New Equipment vs Retrofits
The fact that these (FFR) controls will interact with a wide variety of incumbent controls makes
retrofitting challenging. The reality is that each OEM will need to evaluate what is possible for
each platform (and maybe even each model). Much like the HVDC discussion below, only the
OEMs will really know what is possible. It is by no means guaranteed that they will know what
is possible without making specific (and expensive) investigation within house. Don’t assume
that the OEMs “just know” this stuff, they probably don’t. That has serious implications for
either retroactive requirements or new A/S inducements: the owners of existing wind plants
will not change their equipment, unless the incentive (carrot or stick) is big. High penalties for
failure to respond to retroactive requirements will result in howls of protest.
3.2.6.3! Costs
The capital cost to wind plant owners for inclusion of inertia-based FFR controls (and PFR)
capability in a new plant is expected to be on the order of less than 1% of the capital cost of
the overall project. While this is a small incremental cost, it is important to note that margins
and returns on investment in electric power markets are measured in quite small fractions of
the overall initial capital cost of a project. A less than 1% increase in the capital cost of project
might reasonably be regarded as a substantial sum by the investors.
As noted above, retrofits of inertia-based FFR controls to existing plants are more expensive
than inclusion in new plants. While capital costs for retrofitting the capability in wind plants
vary depending on the vintage and type of technology, a rule of thumb is that the capital cost
for retrofits is on the order of less than 2% to possibly much greater for some OEMs. Again,
this might be a small incremental cost, but in the case of mandatory retrofits, this incremental
cost has not been built into the economic evaluation, justification, and financing for a project,
and so the economic consequence to the plant owner is more significant. It is noted that plant
owners in ERCOT, when faced with a mandatory retroactive requirement (for PFR capability),
were reportedly very unhappy.

3.2.7! Synopsis of Wind Turbine FFR

These are short answers to the questions from AEMO RFP; detailed discussions are provided
in the preceding text.
How fast can it respond? Activation starts with ones of cycles of triggering at the turbine
level, and power can rise on the order of 20-30%/second.

How much can it inject or absorb? Incremental power is generally limited to 10% of pre-
disturbance power.
How long can it sustain this response? Below rated wind speed, but near rated power, the
maximum available arresting energy is on the order of 1 MW-sec/MW; this maximum available
arresting energy drops roughly linearly with power level.
Is there a particular ramp or shape to the response? Ramp rate is limited to 20-30%/second.

56!
! !
!
What determines each of these limits? Is there potential for it to be adapted? Mechanical
stresses and aero-mechanical stability. Pushing further might be possible, but could also
cause loss-of-life in turbines.
Where, when and how have these capabilities been demonstrated? Various OEMs,
including, at least, GE, Senvion, Enercon and Siemens have demonstrated various inertia-
based FFR controls. Tests have been run in the US and Canada, at least.
What is the impact on the resource from responding in this manner? Provide a
quantitative estimate of any immediate (short run marginal) cost. For designs are presently
offered, marginal costs are small. There is nearly, but not quite, zero lost energy production.
And, generally available designs are intended to not use up equipment life.
How mature is this technology? Is there potential for these characteristics to improve in
the future? To what degree, and over what timeframe? First generation offerings have been
available for several years from selected OEMs. There is considerable interest now and new
requirements are emerging rapidly. More options and more field data will almost certainly be
available in the near future.
How is the ability to deliver a fast frequency response service affected by nearby power
system disturbances, such as transmission faults? Will the FFR service still be available in
the period following a transmission fault? How could this be managed? Limitations are
mostly systemic: ability to push extra power into the grid following disturbances is often
voltage dependent. In weak grid controls, active power injection is often slowed or suppressed
to allow for the grid to recover. This will be a serious concern for this application under
conditions when loss of infeed is caused by a fault and trip event. It applies to all FFR
technologies. A somewhat separate consideration is how fast the equipment can to resume
power injection on AC voltage recovery. This is highly dependent on the state of the DC voltage.
This is a design consideration for all FFR and will affect all the inverter-based technologies
discussed in this report. , In general, technologies that resist changes in the DC bus voltage on
the inverters (i.e. the DC bus is “stiff”), like batteries, will tend to be able to recover the fastest.
(This reflected in the subsequent discussions and in the speed notes in the summary tables.)
What is the potential for false or spurious triggering? How is this managed? This is
discussed in Section 1, and is common to all the available technologies.
Are there any other important considerations for the use of this technology to provide a
fast frequency response service, to mitigate high RoCoF? See discussion.

3.3! Battery Energy Storage

Batteries convert chemical energy into electrical energy. An electrolyte allows ions to flow
between the anode and cathode, while a DC electrical current feeds an external circuit.
Batteries and hybrid battery systems (e.g., battery/supercapacitors) have the potential to
provide a useful fast frequency response to mitigate high RoCoF. Some batteries have an
extremely fast response time (e.g., lithium and advanced lead acid) and the technologies range

57!
! !
!
from extremely mature (lead-acid) to less mature (flow). This report focuses on three promising
technologies for FFR: lithium ion, flow, and advanced lead-acid.

In the last several years, batteries have been increasingly deployed in utility applications,
through subsidies, mandates, and especially as demonstration projects for utilities and
governments to better understand the capabilities and limitations of the technology.
Applications include the provision of energy, capacity, ancillary services, transmission and
distribution upgrade deferral, demand charge reduction and backup power. Depending on the
application, different characteristics of energy, power, duration, speed, etc. may be needed.
Batteries can provide value to participants across the power system spectrum. For example,
storage can support the grid to provide capacity to meet system peak, while demand charge
management may allow an individual customer to reduce his individual demand peak. Today,
there are commercial opportunities for batteries in providing fast regulating (secondary)
reserve in markets that pay for performance (e.g., PJM in the US). Batteries are also
commercially viable in some applications and rate structures where it makes sense to help
manage energy bills with existing demand charges and time-of-use rates. In 2013 SBC
reported about 750 MW of large batteries deployed in the utility sector42. Today the US
Department of Energy (DOE) Energy Storage Database lists over 731 battery energy storage
systems (BESS) projects (including Lithium, lead-acid, flow, and sodium technologies) with
1720 MW of total capacity world-wide43.

BESSs are comprised of a power conversion system (PCS) and a battery. Battery cells are
stacked to make up battery modules of the appropriate voltage. These modules may be
controlled with a battery management system (BMS) to protect the battery from over-
charging, over-discharging, and thermal damage. The PCS has a bidirectional inverter that
converts DC output from one or more strings of batteries to AC output and vice versa. For
example, in a 8 MW (32 MWH) BESS in Tehachapi in Southern California Edison, 56 battery cells
(60 Wh, 3.7 V) make up each 3.2 kWh module (52V). Eighteen modules are in each rack and
there are 151 racks in each string of the BESS. Charging/discharging voltages of the strings
are at 760-1050 VDC. Each 2 MW (8 MWH) string has 8 BMSs and one controller. There are two
inverter lineups in each 4 MW PCS which then connects to a 12.47 kVAC grid. The two PCSs are
operated by a master controller which will send real and reactive power commands and may
provide different modes of operation.

Speed of Response

Batteries are not limited by the chemical response time (mass transfer dynamics of ion
migration and depletion), but by the inverter and controls response times. Additionally, the
location of the fast controls, critical sensors, and actuators will have the largest impact on
time because plant level control loops are slow compared to the PCS. The plant level response
time consists of plant level sensing, communication and dispatch to individual inverters.

42!SBC!Energy!Institute,!“Electricity!Storage”,!September!2013.!
43!http://www.energystorageexchange.org/!accessed!July!17,!2016.!

58!
! !
!
Depending on when the event occurs during the control cycle, battery engineers
conservatively estimate that the response is between 150-400 ms. Because of this, an
extremely fast frequency response is better implemented directly on the inverters. The PCS
regulator control loop may have a tens of ms response time. For example, in Hawaii in the US,
requirements for extremely fast response have led to frequency droop implemented directly
on the inverters so that the system begins corrective action within 10 ms, with a full response
within 50 ms. There are drawbacks of PCSs that are required to act fast: PCSs that must
respond within 20ms are maintained in a hot standby condition that has parasitic losses (2%
of rated power)44.

In addition to the BESS response time, total response time for a BESS FFR would include RoCoF
detection/communication time as discussed in section 2.2.2. Whether a direct event detection
scheme or a local RoCoF detection is used, the fastest options for this
detection/communication time appear to be on the order of two cycles. The BESS options
discussed in detail below have activation and full response time on the order of cycles, making
this a potential option for providing arresting energy within the 250 ms needed.

3.3.1! Lithium Batteries

Lithium-ion and lithium polymer-type batteries use lithiated metal oxide as the cathode and
carbon as the anode, with lithium salts as the electrolyte. There are various chemistries
including lithium iron phosphate, lithium manganese oxide, lithium nickel manganese cobalt
oxide, lithium titanate, and lithium nickel cobalt aluminum.
They are widely used in consumer electronics and hybrid/electric vehicles. They are also
increasingly used for utility applications, especially to provide secondary reserves (frequency
regulation) in markets that pay for performance (e.g., PJM in the US) and they make up the
bulk of BESS chemistries being deployed today. To give a sense of lithium battery deployment,
the DOE Global Energy Storage database, that catalogues storage projects around the world,
lists 430 operational projects totaling 1149 MW of lithium BESS installations worldwide.
Low internal impedance leads to high efficiencies (90% and greater). They provide high energy
and power density. Lithium batteries have good cycle life (depends on depth of discharge, and
while typically 1000-10,000 cycles, can be up to several 100,000s cycles), and beyond that,
end of life is marked by degraded performance due to increased internal impedance or
decreased capacity. They also have relatively low self-discharge. Disadvantages are cost and
need for battery management systems. High battery temperatures negatively impact cycle
life so thermal derating in the battery management system limits charge and discharge power
limits based on battery temperature. For example, a lithium BESS may have a
charge/discharge limit equivalent to the power rating of the BESS when cell temperatures are

44!EPRI!and!the!US!Department!of!Energy,!“EPRIXDOE!Handbook!of!Energy!Storage!for!Transmission!and!Distribution!
Applications”,!1001834,!Palo!Alto,!CA,!December!2003.!

59!
! !
!
within limits. At high temperatures, charge/discharge limits may decrease to half of the power
rating. Expected lifetime is 5-15 years.

How fast can it respond? Lithium BESS response times (TActivate+9TActivateFully) are reported to
be about 10-20 ms45,46,47. Documented response times (see below) for tests of lithium BESSs in
the field show tens of ms to hundreds of ms, but we note that these systems were not designed
for the extremely fast responses needed in South Australia. Discussions with industry on
projects that may push the envelope on speed indicate that total response time (TTotalResponse)
can be done in a few cycles. The very fast responses reported for batteries in this are, in part,
likely due to the stiffness of the DC bus, as discussed above.
How much can it inject or absorb? Lithium batteries can inject or absorb 100% of their
capacity but this impacts cycle life. A123 (now NEC) reported a 100,000 cycle lifetime at 100%
depth-of-discharge (DOD)48.
How long can it sustain this response? Response speed is not a driver for thermal stress,
making lithium batteries well-suited for FFR. However, for a system designed for high power,
the magnitude and duration of response over the time frame of hours may lead to constraints
due to internal resistance and mass transfer dynamics that may limit longer, sustained
response. Limitations on response and proper BESS design and cooling capacity manage
various applications’ needs. But for the timeframe of FFR, which may require sustained
response for up to 15 minutes, lithium BESSs are suitable.
Is there a particular ramp or shape to the response? No.
What determines each of these limits? Is there potential for it to be adapted? The
dominating factor of response time of the lithium BESS is the PCS. The battery cell response
time is nearly instantaneous (tens of microseconds). There are tradeoffs in lifetime with DOD
and sustain of response. Managing thermal impacts means there may be impacts in terms of
efficiency for parasitic loads and in terms of capital cost with increased cooling capacity for
higher performance.
Where, when and how have these capabilities been demonstrated? Over one hundred
projects with total capacity of over 800 MW of lithium BESSs have been installed at the utility-
scale for frequency regulation services49. While frequency regulation calls for responses on

45!SBC!Energy!Institute,!“Electricity!Storage”,!September!2013.!
46!C.!Vartanian,!A!23,!Grid0Stability0Battery0Systems0for0Renewable0Energy0Success,!www.neces.com!
47!HDR!Engineering,!Inc.,!“Update!to!Energy!Storage!Screening!Study!for!Integrating!Variable!Energy!Resources!within!the!
PacifiCorp!System,”!Salt!Lake!City,!UT,!July!9,!2014.!
48!HDR!Engineering,!Inc.,!“Update!to!Energy!Storage!Screening!Study!for!Integrating!Variable!Energy!Resources!within!the!
PacifiCorp!System,”!Salt!Lake!City,!UT,!July!9,!2014.!
49!http://www.energystorageexchange.org/!accessed!July!17,!2016.!

60!
! !
!
the order of seconds instead of tens to hundreds of milliseconds, it gives an indication of the
level of deployment of lithium BESSs for fast response.

An example of a lithium-ion BESS providing fast response to stabilize frequency is on the Big
Island of Hawaii in the US. A 1 MW/250 kWh BESS was commissioned at the Hawi wind plant
on the Hawaii Electric Light Company (HELCO) grid in January 2013. On this 180 MW system,
variability at two wind plants has led to strong frequency fluctuations. A GE study led to the
installation of a BESS with funding from the US Department of Energy. Two real-time control
algorithms were developed: one to balance generation and load within 100 ms and another
to smooth strong variations in wind within 200 ms50. Frequency variability has been reduced
by 30-40% with the BESS. The frequency regulation algorithm sets BESS output as a function
of frequency (not RoCoF). Response times were fast, comprised of the site dispatch controller’s
nominal computation time of 100 ms and 100 ms for the BESS’s output to reach 90% of the
command. A test of this latter capability is shown in Figure 22.

Another example, from the first Chilean BESS used to provide primary frequency response,
was installed in 2009. This 12 MW/4 MWh lithium nanophosphate battery system controls
output as a function of frequency (not RoCoF): outside of a +/- 0.3 Hz deadband it responds to
overfrequency with a full lower, and to underfrequency with a full raise. While full power
response times have been as fast as tens of ms, typical latencies are in the hundreds of ms51.
Another example is the 500 kWh BYD lithium-iron phosphate battery in Qatar, with a < 100 ms
response time. This system provides reactive power support, frequency regulation and black
start capabilities52,53.

50!Hawaii!Natural!Energy!Institute,!Development0of0RealGtime0Closed0–loop0Control0Algorithms0for0GridGscale0Battery0Energy0
Storage0Systems,!Aug.!2014.!!
51!Hsieh,!E.!and!Johnson,!R.!“Frequency!Response!from!Autonomous!Battery!Energy!Storage”,!CIGRE!US!National!Committee!
2012!Grid!of!the!Future!Symposium,!Paris,!2012.!!
52!IRENA,!“Case!studies:!battery!storage”,!!
http://www.irena.org/DocumentDownloads/Publications/IRENA_Battery_Storage_case_studies_2015.pdf!!
53!BYD,!http://www.byd.com/energy/ess.html.!

61!
! !
!
 Batteries and National Grid’s Enhanced Frequency Response
The U.K.’s National Grid has established an Enhanced Frequency Response (EFR) service. A
tender for 200 MW of EFR was issued in July 2016. EFR acts when frequency is outside of a
49.5 – 50.5 Hz deadband. The time for detection, signaling, and delivering the MW output
must be no greater than 1 second in total, with the time for detection and signaling to be no
greater than 500 ms. Full output should be sustained for at least 15 minutes. While the
auction is technology-neutral, battery developers are highly interested in providing EFR, as
demonstrated by the 888 MW of battery projects (out of a total of 1.3 GW of projects) that
responded to a call for expressions of interest. Renewable Energy Systems (RES) is already
working with National Grid outside of this auction to deploy 20 MW of battery storage that
can provide sub-second frequency response. This is expected to be operational by the end
of 2017.

Firm Frequency Response is divided into low frequency primary (full output within 10 s and
sustained for another 20 s), low frequency secondary (full output within 30 s and sustained
for 30 min), and high frequency (full output within 10 s and sustained indefinitely). The largest
installed battery system in the UK is the Smarter Network Storage project 6 MW (10 MWh)
which uses Samsung SDI batteries and S&C Electric inverter systems. This system relieves
thermal overhead line constraints and also transformer constraints at the Leighton Buzzard
distribution substation. It provides peak shaving, frequency response, reserves, and
arbitrage. The graphs below show (top) testing of the system providing dynamic firm
frequency response and (bottom) the system providing static firm frequency response.

62!
! !
!
What is the impact on the resource from responding in this manner? Provide a
quantitative estimate of any immediate (short run marginal) cost. For a lithium BESS to
provide FFR with a sustain of 15 minutes or less, there is no impact on the resource lifetime or
increase in short run marginal cost. That being said, cycle life (typically 1,000-10,000 cycles) is
an issue for lithium BESS. Capacity is typically degraded in 5-10 years and while those batteries
may be used in other applications, they may need to be replaced to continue to provide FFR.
How mature is this technology? Is there potential for these characteristics to improve in
the future? To what degree, and over what timeframe? Lithium batteries are not as mature
as lead-acid batteries, but deployment is rapid. Their development has been driven by
consumer electronics and they are the technology of choice for electric vehicles. Of the BESS
technologies currently being deployed in utility applications, lithium dominates the market.
Their relative immaturity combined with wide deployment means that costs are dropping
quickly and have a great potential to decrease further. With increased interest in BESS for FFR
in Ireland, ERCOT, the NEM and other jurisdictions, it is likely that speed and lifetime can
improve as well in the near future.
What is the potential for false or spurious triggering? How is this managed? As with the
other inverter-based technologies in this section, this depends on the RoCoF detection scheme
(see section 2.2.2.2).
Are there any other important considerations for the use of this technology to provide a
fast frequency response service, to mitigate high RoCoF? Safety has been an issue with
lithium-ion batteries, with overheating being a potential problem. LiFePO4 is one of the two
commercial lithium chemistries that is not subject to thermal runaway and therefore
considered by some to be safer.

63!
! !
!
 Figure 22 Performance of HELCO BESS using frequency regulation algorithm54.
The black segments show battery power, grid frequency and wind power when the BESS is inactive; the
red segments show the same parameters when the BESS is active.

3.3.2! Flow Batteries

Flow batteries are similar to fuel cells. They have active materials in solution in the electrolyte
at all times. Two separate electrolytes are separated by a membrane which allows ion

54!Hawaii!Natural!Energy!Institute.!Development0of0RealGtime0ClosedGloop0Control0Algorithms0for0GridGscale0Battery0Energy0Storage0
Systems,!Aug.!2014.!

64!
! !
!
permeability, while current flows through an external circuit. The electrolyte is pumped past
the membrane and external tanks hold some amount of electrolyte, which determines the
energy capacity rating of the battery. A commercial flow battery technology is the vanadium
reduction and oxidation (redox) battery (VRB) which is based on different ionic states of
vanadium. A proton exchange membrane allows ions to flow and a vanadium/sulfuric acid
mixture is used as the electrolyte for both sides of the cell. Other chemistries include zinc
bromine, hydrogen bromine and zinc nickel oxide.
The DOE Global Energy Storage database lists 67 operational flow BESS projects worldwide,
totaling 74 MW55. An advantage of flow batteries is that power and energy can be scaled
independently: power ratings depend on the surface area of the electrodes while energy
capacity depends on the volume of electrolyte in the tanks. This makes flow batteries
attractive for long duration storage. Flow batteries also have the ability to use 100% of the
stored energy. They are tolerant of over-charging, over-discharging and wide temperature
ranges. The electrolytes are stored in separate tanks so self-discharge is minimal56.

Another advantage is the long cycle life and lack of degradation of power or energy ratings
over the lifetime of the battery. This is because cycling a flow battery does not cause a physical
change in the electrode, as it does in most other battery technologies. Cycle life is thus
independent of depth of discharge, in contrast to other batteries. Lifetime is limited by the cell
stack and other components (tank, power electronics)57.
Disadvantages include the more complex nature of the system. Flow batteries do not have
high efficiencies. VRB efficiencies are on the order of 60-85%58.
How fast can it respond? If the VRB is in a ‘ready’ state, with its stacks primed with reactants,
they can increase from zero to full output (TActivate+9 TActivateFully) within a few milliseconds59.
Short duration discharges can be responded to without even running the pumps60. Response
time is thus limited by the controls and communications equipment, similar to other batteries.
For example, Prudent Energy’s VRB system claims 100,000 cycles lifetime at full rated depth
of discharge and a step response from charge to discharge of less than 50 ms61. UniEnergy
Technologies’ VRB system claims unlimited cycles over the 20 year life of the system, with

55!http://www.energystorageexchange.org/!accessed!July!17,!2016.!
56!UET!specifies!less!than!2%!selfXdischarge!due!to!residual!electrolyte!left!in!stacks,!with!no!discharge!of!energy!remaining!in!
electrolyte!tanks!over!time.!https://www.nwppa.org/wpXcontent/uploads/UETXTheXOnlyXMegawattXScaleXContainerizedXFlowXBatteryX
NWPPAXpresentationX05X17X16.pdf!!
57!Akhil,!A.,!Huff,!G.,!Currier,!A.,!Kaun,!B.,!Rastler,!D.,!Chen,!S.!Cotter,!A.,!Bradshaw,!D.,!Gauntlett,!W.,!“DOE/EPRI!2013!Electricity!
Storage!Handbook!in!Collaboration!with!NRECA”,!SAND2013X5131,!July!2013.!
58!SBC!Energy!Institute,!“Electricity!Storage”,!September!2013.!
59!Akhil,!A.,!Huff,!G.,!Currier,!A.,!Kaun,!B.,!Rastler,!D.,!Chen,!S.!Cotter,!A.,!Bradshaw,!D.,!Gauntlett,!W.,!“DOE/EPRI!2013!Electricity!
Storage!Handbook!in!Collaboration!with!NRECA”,!SAND2013X5131,!July!2013.!
60!ibid.!
61!Prudent!Energy!2011!brochure.!

65!
! !
!
100% available state-of-charge and a response time (TActivate+9 TActivateFully) of less than 100
ms62.
How much can it inject or absorb? Flow batteries can easily inject or absorb 100% of their
capacity. Additionally the cell stack can produce three times rated power output if the state
of charge is between 50 and 80 percent63.
How long can it sustain this response? An advantage of flow batteries is their ability to scale
energy. Flow batteries can sustain their response for as long as they have capacity (which is
a function of how much electrolyte is in the tanks) and there is no limitation on capacity.
Is there a particular ramp or shape to the response? No.
What determines each of these limits? Is there potential for it to be adapted? Similar to
other BESSs, technical limitations on speed are determined by the PCS. The total response time
is also determined by the RoCoF detection and communications time. The flow battery itself is
not the limiting factor.
Where, when and how have these capabilities been demonstrated? Flow batteries have
been deployed around the world, especially in Asia, and typically for demonstration or pilot
purposes. There are far fewer flow batteries providing fast regulation or other fast services,
compared to lithium. Of the 67 operational flow BESS projects in the DOE Global Energy
Storage database, six projects totaling 5 MW were providing frequency regulation services64.
In Japan, Hokkaido Electric Power Company installed Sumitomo Electric Industries’ 15
MW/60MWh vanadium redox system to help manage wind and solar variability6566. In
Shenyang, China, Guodian Longyuan Windpower Company installed a Dalian Rongke Power’s
5 MW/10 MWh vanadium redox battery system at the Faku wind plant to smooth wind output,
regulate frequency and support voltage6768. In Zhangbei, China, the China Electric Power
Research Institute installed Prudent Energy’s 500 kW/1MWh vanadium redox battery system
at a wind and PV plant to provide fast balancing services6970. In the US in April 2012, Prudent
Energy installed a 0.6 MW behind-the-meter vanadium redox battery system in a California
commercial process plant for demand charge reduction and to manage energy bills. In
Pullman, WA in June 2015, Avista Utilities commissioned a 1 MW/3.2MWh vanadium redox

62!http://www.cesa.org/assets/Uploads/ESTAPXSlidesX6.19.14.pdf!!
63!Mears,!L.,!Gotschall,!H.,!Key!T.,!Kamath,!H.,!“EPRIXDOE!Handbook!of!Energy!Storage!for!Transmission!and!Distribution!
Applications”,!EPRI,!Palo!Alto,!CA,!2003.!!
64!http://www.energystorageexchange.org/!accessed!July!17,!2016.!
65!http://www.energystorageexchange.org/projects/2043!accessed!July!17,!2016.!
66!http://renewables.seenews.com/news/japanXsXhepcoXseiXkickXoffX15XmwXbatteryXsystemXverificationX507909!!
67!http://www.energystorageexchange.org/projects/1624!accessed!July!17,!2016.!
68!http://www.rongkepower.com/index.php/article/show/id/140/language/en!!
69!http://www.pdenergy.com/pdfs/CEPRIProjectXFactSheet053112XFINAL.pdf!!
70!http://www.energystorageexchange.org/projects/159!accessed!July!17,!2016.!

66!
! !
!
battery for load shifting, frequency regulation and conservation voltage regulation on a
distribution feeder.

What is the impact on the resource from responding in this manner? Provide a
quantitative estimate of any immediate (short run marginal) cost. There is no impact on
the resource lifetime or increase in short run marginal cost for the aforementioned level of
performance.
How mature is this technology? Is there potential for these characteristics to improve in
the future? To what degree, and over what timeframe? Flow batteries are not as mature as
lead-acid batteries and deployment is slower than for lithium batteries. While they have been
around longer than lithium, they are used infrequently and don’t really have applications in
transportation or consumer electronics, which might decrease costs and increase
deployment.

What is the potential for false or spurious triggering? How is this managed? As with the
other inverter-based technologies in this section, this depends on the RoCoF detection scheme
(see section 2.2.2.2).
Are there any other important considerations for the use of this technology to provide a
fast frequency response service, to mitigate high RoCoF? Flow batteries have relatively low
energy and power densities and therefore require more space than other BESSs. The balance
of plant for flow batteries is also a bit more complicated.

3.3.3! Lead-acid Batteries

Lead-acid batteries are the oldest, most mature, commercial battery storage technology. They
are used worldwide to start automobile and boat engines. Lead-acid batteries are the
cheapest in terms of energy storage. While they can be discharged quickly, charging is slow,
and aging leads to sulfation of the negative electrode, degrading performance.
Advanced lead-acid carbon batteries add carbon to the negative plate in various ways to
increase power performance and handle partial states of charge. Some variants essentially
add a supercapacitor to the lead-acid battery, such as the one shown in Figure 23, thus
allowing the battery to provide an extremely fast response while also allowing for long
duration time.

67!
! !
!
 Figure 23 The Ecoult UltraBattery (hybrid supercapacitor and a lead-acid battery)
Advanced lead-acid carbon batteries can provide a high charge and discharge rate similar to
that of lithium-ion batteries. Manufacturers report high cycle life (e.g., 1000-1600) with deep-
discharges71.

How fast can it respond? Figure 24 and Figure 25 show the fast response time, TActivate+9
TActivateFully, (<200 ms for full response) of a 5 MW (2.5 MWH) conventional lead-acid BESS in
Vernon, CA that was installed in 1995 to provide backup power for critical processes at an
industrial facility. Those tests are of considerable interest for AEMO, as they represent
successful separation from the main grid (1st figure), and successful resynchronization with
the main grid (2nd figure). The separated system has no synchronous generation of any kind
running during this test. The BESS control here is responsible for all frequency control for the
island. Consequently, while this demonstrates the capability of a BESS to respond, the
example is not identically equal to FFR. Zero synchronous generation operation is discussed
further in section 4.5.

71!Akhil,!A.,!Huff,!G.,!Currier,!A.,!Kaun,!B.,!Rastler,!D.,!Chen,!S.!Cotter,!A.,!Bradshaw,!D.,!Gauntlett,!W.,!“DOE/EPRI!2013!Electricity!
Storage!Handbook!in!Collaboration!with!NRECA”,!SAND2013X5131,!July!2013.!

68!
! !
!
 Figure 24 BESS supporting a trip to island event72.
The x-axis is time, with each square representing 1 second.

72!Miller,!N.!W.,!et!al.,!“Design!and!Commission!of!a!5!MVA,!2.5!MWH!Battery!Energy!Storage!System,!IEEE!SPM!1996.!!

69!
! !
!
 Figure 25 BESS enabling resynchronization of island to main grid73.

The x-axis is time, with each square represens 1 second.

As with other BESS chemistries, total response times for advanced lead-acid BESSs are the
sum of RoCoF detection time, communication time and response time of technology.
Advanced lead-acid BESSs such as the UltraBattery (supercapacitor/battery) shown above are
reported to respond in about 40 ms74.

Figure 26 shows the performance of an advanced lead-acid BESS from Xtreme Power that
was installed in ERCOT in 2013. Response time for this BESS to detect the issue (under-
frequency in this application, not RoCoF), and fully respond (TTotalResponse) is within 92 ms75. (This

73!Miller,!N.!W.,!et!al.,!“Design!and!Commission!of!a!5!MVA,!2.5!MWH!Battery!Energy!Storage!System,!IEEE!SPM!1996.!!
74!HDR!Engineering,!Inc.,!“Update!to!Energy!Storage!Screening!Study!for!Integrating!Variable!Energy!Resources!within!the!
PacifiCorp!System,”!Salt!Lake!City,!UT,!July!9,!2014.!
75!Duke!Energy,!“Technology!Performance!Report:!Duke!Energy!Notrees!Wind!Storage!Demonstration!Project:!2015!Final!Report”,!
Nov.!2015.!https://www.smartgrid.gov/files/OE0000195_Duke_FinalRep_2015.pdf!

70!
! !
!
response is based on local measurements, and is subject to the concerns raised above in
Section 2.2.2)

Figure 26 Notrees providing fast responsive reserves in ERCOT on Nov 1, 201376.

Figure 27 BESS arrests under-frequency deviation in ERCOT77.

76!Ibid.!
77!Ibid.!

71!
! !
!
Left axis is energy in kW for the XP Power BESS (blue trace) and ERCOT AGC signal (red trace); right axis is frequency
in Hz for ERCOT (green trace).

In Figure 27, the BESS responds nearly instantly to frequency dropping below 59.91 Hz. The
BESS response gives the AGC time to ramp secondary reserves up.

Figure 28 BESS arrests extreme frequency deviation in ERCOT78.

Left axis is energy in kW for the XP Power BESS (blue trace) and ERCOT AGC signal (red trace); right axis
is frequency in Hz for ERCOT (green trace).

Figure 28 shows another event with greater time resolution demonstrating the ability of the
BESS to trigger and provide full response for under-frequency within cycles.
How much can it inject or absorb? Typical deep discharge lead-acid batteries may have a
depth of discharge (DOD) of 60-80% but there are 100% DOD batteries developed for utility
applications: Xtreme Power reported 1000 cycles at 100% DOD79. Ecoult’s UltraBattery
reported 9000 cycles at 100% DOD. Higher DOD negatively impacts cycle lifetime.
How long can it sustain this response? Lead-acid batteries can sustain a response for hours,
depending on energy rating.
Is there a particular ramp or shape to the response? No.
What determines each of these limits? Is there potential for it to be adapted? The
development of advanced lead-acid batteries that combine supercapacitors with batteries or
alter the negative electrode give a fast and sustained response with 100% DOD. There are

78!Ibid.!
79!HDR!Engineering,!Inc.,!“Update!to!Energy!Storage!Screening!Study!for!Integrating!Variable!Energy!Resources!within!the!
PacifiCorp!System,”!Salt!Lake!City,!UT,!July!9,!2014.!

72!
! !
!
also hybrid technologies that combine batteries with flywheels to give a similar high
performance.

Where, when and how have these capabilities been demonstrated? The DOE Global Storage
database lists 83 operational lead-acid BESS projects totaling 110 MW80. Fifteen of these are
used for frequency regulation applications. Of these BESSes, 30 projects totaling 70 MW utilize
advanced lead-acid batteries.
Before bankruptcy, Xtreme Power deployed several of their advanced lead-acid PowerCell
systems. A number of these were installed in Hawaii to provide ramp rate control, voltage
support, wind/solar smoothing, and/or reserves. For example, the 10 MW (20 MWH) BESS in
Maui integrates a wind plant and the 1.1 MW (0.5 MWH) BESS in Lanai in 2011 integrates a PV
plant. The Xtreme Power BESS shown in figures above was extensively tested and
documented. This BESS was installed in the Notrees wind plant and provided fast frequency
regulation in ERCOT81. The batteries, which had been commissioned Jan 2013, were no longer
operating at full capacity by Oct 2015. Operations of that BESS had to shift to reflect the
degraded performance and the batteries are being replaced with lithium ion in 2016.
A 3 MW East Penn UltraBattery (supercapacitor/battery) was commissioned in 2012 to provide
frequency regulation services in PJM82.
What is the impact on the resource from responding in this manner? Provide a
quantitative estimate of any immediate (short run marginal) cost. There is not an impact
from the speed or the sustain of the performance. There is a tradeoff between DOD and cycle
lifetime, as with most batteries (not flow batteries).
How mature is this technology? Is there potential for these characteristics to improve in
the future? To what degree, and over what timeframe? Lead-acid batteries are a very
mature technology but their adaptation to advanced lead-acid batteries to provide fast
services in the utility sector is relatively new, giving them potential for further improvement in
the near term.
What is the potential for false or spurious triggering? How is this managed? As with the
other inverter-based technologies in this section, this depends on the RoCoF detection scheme
(see section 2.2.2.2).
Are there any other important considerations for the use of this technology to provide a
fast frequency response service, to mitigate high RoCoF? Lead is toxic, but recycling is well-
known and understood.

80!http://www.energystorageexchange.org/!accessed!July!17,!2016.!!
81!Duke!Energy,!“Technology!Performance!Report:!Duke!Energy!Notrees!Wind!Storage!Demonstration!Project:!2015!Final!Report”,!
Nov.!2015.!https://www.smartgrid.gov/files/OE0000195_Duke_FinalRep_2015.pdf!!
82!Akhil,!A.,!Huff,!G.,!Currier,!A.,!Kaun,!B.,!Rastler,!D.,!Chen,!S.!Cotter,!A.,!Bradshaw,!D.,!Gauntlett,!W.,!“DOE/EPRI!2013!Electricity!
Storage!Handbook!in!Collaboration!with!NRECA”,!SAND2013X5131,!July!2013.!

73!
! !
!
3.4! Flywheel Energy Storage
3.4.1! Flywheel Energy Storage Systems (FESS) with Inverter Interfaces
The relatively new generation of inverter interfaced flywheels are a commercially available
technology with advantages of a high power density and high cycle life. Disadvantages
include a low energy density.
These devices have two types of electrical topology that closely resemble the two main classes
of utility-scale wind turbines on the market today: devices that run at very high speeds and
which interface to the grid through full-converters, much like “type 4” wind turbines, and
devices that run close to synchronous speed, but which have significant speed deviation from
synchronous speed enabled by a double-fed topology that is much like “type 3” double-fed
wind turbines.
Deployments of flywheels, as measured in both count of projects and total MW/MW-sec
rating, are relatively small compared to other commercial storage technologies. Essentially all
of today’s commercially available flywheels are high-speed and use full converters. They are
ideal for frequency regulation and other fast response services due to their fast response time
and high power to energy ratio. High power flywheels can recharge in seconds83. They have
very high efficiencies of over 90%84. Energy flywheels (as opposed to power flywheels) are
being designed for longer durations with advanced technologies to reduce stand-by losses
and increase round-trip efficiency. Cycle life is typically estimated at over 100,000 cycles85,
although newer configurations report 175,000 cycles86. There is virtually no wear-and-tear
from cycling. The DOE Global Storage database lists 40 operational flywheel energy storage
projects totaling 930 MW87. Twenty-one of these are used for frequency regulation
applications. The following responses apply to these full-converter, high speed flywheels,
unless otherwise noted.
How fast can it respond? In 2013, Sandia/EPRI/DOE reported for power flywheels: very fast
response times (TActivate+9 TActivateFully) of 4 ms or less. Esave’s flywheel solution claims a
response time (TActivate+9TActivateFully) in as little as 10 ms88.
How much can it inject or absorb? 100% of capacity.
How long can it sustain this response? Typical FESS deployment has been designed to
provide power for frequency regulation with sustain times on the order of 15 minutes.
However, flywheels can also be designed to provide energy with longer durations, on the order

83!Ibid.!
84!Ibid.!
85!Ibid.!
86!http://energystorage.org/energyXstorage/technologies/flywheels!!
87!http://www.energystorageexchange.org/!accessed!July!17,!2016.!!
88!http://www.esavecorp.com/?page_id=3873!!

74!
! !
!
of hours. An Amber Kinetics project in San Diego Gas & Electric is demonstrating a four hour
flywheel89.

Is there a particular ramp or shape to the response? No.

What determines each of these limits? Is there potential for it to be adapted? FESSs are
incredibly fast. The technology response time, capacity that can be utilized, and sustain time
are all suitable for AEMO needs. The constraining factor for FESS will be RoCoF detection and
communication time.
Where, when and how have these capabilities been demonstrated?
In the US, Beacon Power installed a 20 MW (15 minute) flywheel in 2011 to provide frequency
regulation to the NYISO market. Spindle Grid Regulation, who now owns Beacon Power, also
installed a 20 MW (15 minute) flywheel in the PJM market to provide fast frequency regulation
services. A 23 MW flywheel is in operation in Okinawa Power, providing frequency regulation90.
A flywheel-battery hybrid with two 160kW flywheels and 240 kW (80kWh) of batteries was
installed in Ireland by Schwungrad Energie to demonstrate various fast responses that could
be provided in EirGrid’s DS3 program. The power conversion system was designed for
response times (TActivate+9TActivateFully) of less than 20 ms91 so that the system could provide a
fast frequency response and limit RoCoF.
Industrial flywheels have long been used for specialized applications. For example, GE
provided a 5MW, 40 ton flywheel on the dragline at the Usabelli Coal Mine in Alaska, installed
in 197892. It uses an early (silcomatic) power converter. It smoothes out the cyclic loading and
regeneration of this mining load that had previously been highly disruptive to the rural grid to
which it is attached.
The largest flywheels in the world are used in the high energy physics research communities
to provide large amounts of power for very short periods (seconds) of time. For example, the
Max-Planck Institute in Germany has been using a FESS since 1987 with three units that have
capacities of 155 MW (9.7 second), 124 MW (4.4 second), and 108 MW (6.7 second)93. But these
installations are not part of utility grid operation for discharge, as they are synchronous
machines that depend on substantial deviations of terminal frequency to extract the energy
from the rotor. (more below, in the next subsection).

What is the impact on the resource from responding in this manner? Provide a
quantitative estimate of any immediate (short run marginal) cost. There is no impact on

89!Akhil,!A.,!Huff,!G.,!Currier,!A.,!Kaun,!B.,!Rastler,!D.,!Chen,!S.!Cotter,!A.,!Bradshaw,!D.,!Gauntlett,!W.,!“DOE/EPRI!2013!Electricity!
Storage!Handbook!in!Collaboration!with!NRECA”,!SAND2013X5131,!July!2013.!
90!Ibid.!
91!http://www.energyXstorage.news/news/flywheelXbatteryXhybridXsystemXinstalledXinXireland!!
92!http://energyXalaska.wikidot.com/usibelliXflywheel!!!!
93!http://www.energystorageexchange.org/projects/918!accessed!July!17,!2016.!

75!
! !
!
the flywheel from providing fast response or 100% response. Longer sustain may require
design for energy, not power, or hybridization with a battery.

How mature is this technology? Is there potential for these characteristics to improve in
the future? To what degree, and over what timeframe? Flywheels have been around for a
long time and are commercial, but they have not been widely deployed. The technology is
moderately mature and there is potential for improvements especially as energy flywheels are
further developed and increase the duration of the storage capacity.
How is the ability to deliver a fast frequency response service affected by nearby power
system disturbances, such as transmission faults? Will the FFR service still be available in
the period following a transmission fault? How could this be managed? This service should
still be available following a transmission fault. The issue of DC bus stiffness discussed above
applies to full inverter flywheels and double fed machines. The maintenance and recovery of
DC bus voltage will depend on design, particularly control of the DC chopper for full converter
flywheels. Good performance should be possible.
What is the potential for false or spurious triggering? How is this managed? As with the
other inverter-based technologies in this section, this depends on the RoCoF detection scheme
(see section 2.2.2.2).
Are there any other important considerations for the use of this technology to provide a
fast frequency response service, to mitigate high RoCoF? No.

3.4.2! Flywheel Energy Storage Systems with Large Synchronous and Near-synchronous
Machines
As noted above, there is a class of flywheels that use double-fed topology. These are
discussed along with synchronous devices in this subsection. In general, this document is
dedicated to discussion of technologies other than conventional synchronous generation and
transmission alternatives to address the challenge of managing high RoCoF and severe
frequency excursions. However, there is one emerging variation on traditional synchronous
condensers that warrants discussion here: synchronous condensers with enhanced inertia.
Most recently large (200MVA) class synchronous condensers with inertia constants of around
8 MW-sec/MVA have been built to support frequency, voltage and short circuit strength at the
inverter end of a new HVDC project in Canada. This technology has several attributes that
may be especially attractive for South Australia, and perhaps Tasmania as well. In addition to
providing synchronous inertia, the devices will provide dynamic voltage support and short
circuit strength. [Additional dynamic voltage support would have been welcome during the
28 September 2016 event.18] Capital costs for high inertia synchronous condensers are on the
order of $25US/kW-sec.
There are also hybrid devices, that have been called “rotating stabilisers94”, which operate as
synchronous condensers down to the frequency nadir. At the frequency nadir, rather than

94!“Rotating!Stabiliser”!has!been!tradeXmarked!by!GE.!

76!
! !
!
reaccelerating, they switch to a full converter topology. This is accomplished by disconnecting
the synchronous connection of the stator and feeding it through a full ac-dc-ac converter. In
this second mode they share topology with the class of flywheel systems discussed above in
Section 3.4.1. The rating of the converter is selected to match the desired power-time
relationship needed for the application (more discussion in the next section). The inertia
constant for these devices can be customized, and is reported to be available in range of 20-
40 MW-sec/MW. This suggests that a single installation could maintain South Australia inertia
at the low end present range of operation for conceivable future operation at zero
synchronous generation. However, loss of the majority of the voltage support and short circuit
contribution in this mode might be problematic.
While not exactly commercialized, there is potential to use the variable speed technology that
enables variable-speed pumped storage hydro (VSPSH) as flywheel energy storage system.
We have not discussed VSPSH in this section, with the understanding that longer term energy
storage technologies are out of scope for this report. Further geographic topology for PSH is
probably not good in South Australia, but the double fed technology for VSPSH is applicable
to flywheels on a large scale. VSPSH is a small, but growing base of application. As far as we
can determine, there are on the order of a dozen doubly-fed VSPSH projects built or under
construction in the world. Ratings are mostly on the order of 100-200MW. The GE machines
under construction are on the order of 200MVA machines.
3.4.2.1! Big Flywheel Discussion: Synchronous vs. Variable Speed
Some comparative discussion of large machines with high inertia is helpful.
For the sake of comparison, we introduce 200 MVA machine, with augmented inertia to an
inertia constant H = 8 (at synchronous speed), which results in a total inertia of 1600MW-sec.
This is a big enough inertia to have a noticeable impact on the RoCoF and frequency problems
of concern in South Australia (and Tasmania).
The fundamental physics of a flywheel dictates that, to get power in or out, the speed needs
to change. The energy stored is proportional to the square of the speed (duh!). A linear
approximation is, therefore, that for each 1% change in speed the energy delivered
(consumed) changes by 2%. A synchronous machine is, by definition, constrained to a speed
proportional to the grid frequency. Therefore, the energy that can be extracted is constrained
by the amplitude of the allowable frequency excursion. In laboratory settings, where
synchronous flywheels are used to power plasma devices and particle accelerators, the
machine is “spun up”, and discharges its energy with a decaying AC frequency –
desynchronized from the host grid. Interesting, but of no use to AEMO. For a synchronous
machine in the NEM, the contribution of synchronous inertia (primarily) occurs within the
frequency limits of the grid, i.e. before the UFLS triggers at 49Hz. Since 49Hz is a 2% deviation,
4% of the stored energy is useful: just like ALL the other synchronous generators that might
still be running. The contribution is inherent, instantaneous and doesn’t depend on fast
sensing, communication or actuation. For our numerical example, the arresting energy
delivered is:
MVA*H*((1)2 – (1-dF)2) = 200*8*((1.02)-(.982)) = 200*8*0.04 = 64 MW-sec.
77!
! !
!
With a double-fed arrangement, the machine speed can change over a much wider range.
The enabling technology is the same as that of double—fed (type 3) wind turbine. The power-
electronics provides excitation and enables slip. The converter rating is proportional to the
slip speed. So, for our 200MVA, 1600MW-sec, and example 30MVA (nominal) ac-dc-ac
inverter gives +/- 15% speed range, which means that the energy that can be actually used
is 60% of the stored energy:
MVA*H*((1+dSlip)2 – (1-dSlip)2) = 200*8*((1.152)-(.852)) = 200*8*0.6 = 960 MW-sec.
Which equals the machine rated power delivered for 4.8 seconds: 15 times the energy of the
synchronous machine. To a first approximation, the machine cost is the same (or somewhat
higher), but the power electronics of rating about 15% of the machine, adds cost. With this
arrangement, voltage control may be faster, and maintaining synchronism will be easier (this
device will not be subject to traditional first swing stability constraints, as a synchronous
device will necessarily be. But, aside from higher capital, with this arrangement there are
other draw backs:
(a)! the machine relies on control – so the 3 elements: measure, communicate, activate apply.
This is not synchronous inertia, this is FFR.
(b)! The machine does not contribute to short circuit strength, rather it exacerbates the low
SCR (weak grid; high SNSP) problem.
(c)! It is less widely established technology.
Nevertheless, the scale and agility of the device, coupled with large amount of arresting
energy that could be delivered rapidly appear to make this technology attractive for this
application.

The hybrid “rotating stabilizer” configuration is a significant improvement on some aspects of

the trade-off between synchronous and variable speed. To continue the numerical example,
while in synchronous mode, the device will deliver 4% of the stored inertia energy and all the
other benefits of being a synchronous machine up to the time of the nadir. At the time of the
nadir, the device switches to provide continued primary frequency response, by further
decelerating the rotor. The converter allows for nearly all of the remaining rotational energy
(1536 MW-sec) to be delivered to the system. The power rating of converter dictates the
maximum power that can be delivered. So, for the numerical example, a 15% rated (i.e. 30MVA)
converter, would allow 30MW to be delivered for about 50 seconds.
So, in comparison to the double-fed arrangement, which allows the full MW rating (of the
machine) to be delivered, but which cannot extract all of the stored energy, this arrangement
allows all the rotational energy to be delivered, but at a power level limited by the converter
rating. Table 5 provides a concise comparison of the three topologies. The rotating stabilizer
is normally synchronous, and switches to full-converter at the frequency nadir.

78!
! !
!
 Table 5 Comparison of Available Energy for Flywheel Topologies
Synchronous Double-Fed Full Converter
Usable Energy !F2 2!#max2 : Perated = 100%
(% of ½ "#o2) (~4%) !smax= Perated (~100%)
(~60%)
SIR/SCR contribution yes no no
FFR/PFR No (not controlled) Yes Yes
(Pmax = 100%) (Pmax = Perated )
Q 100% (machine Q 100% (machine Q Perated (converter Q
rating) rating) rating)

Ultimately, it is unlikely that any of these devices will be justified solely on their inertia/FFR
benefit, but rather they could be justified on all three attributes: inertia, dynamic voltage
support and short-circuit contribution. As the discussion illuminates, there are a range of
options available to the designer. These machines will inevitably be customized for the specific
needs of the application, and trade-offs between performance in these 3 aspects, balanced
with cost evaluation of no-load and full-load losses, and capital costs will have to be made
based on detailed and specific requirements. These can only be given as the result of
thorough systemic analysis. One design most emphatically does not “fit all” here.

3.5! Supercapacitor Energy Storage Systems

Supercapacitors, or ultra-capacitors, are electrochemical capacitors. They are similar to
batteries in that ions migrate from one electrode to another through an electrolyte. Unlike
batteries, there is no chemical reaction. Therefore wear-and-tear from cycling is not an issue
and cycling does not impact lifetime. Electrodes are typically made of porous carbon which
has a high surface area, which provides extremely high storage capacity. They have a much
higher power density than other storage technologies discussed here. Because they have very
low internal impedance, they can provide high currents. Key advantages are high cycle life
(100,000 cycles), high power, high efficiency, and fast response times. Supercapacitors can be
charged and discharged rapidly, in 1-10 seconds.
Disadvantages include low energy, cost, and need for power conditioning. Power conditioning
is necessary because capacitor voltage is proportional to the stored charge in the system.
Supercapacitors are used in the transportation industry for regenerative braking, voltage drop
compensation, or for short bursts of power, and in industrial applications as uninterruptible
power supply and also in consumer electronics. Their use in the utility sector is relatively new,

79!
! !
!
and they tend to be used in hybrid applications with batteries or as part of advanced lead-
acid batteries.

How fast can it respond? SBC quotes response times (TActivate+9TActivateFully) of 10-20ms95.
How much can it inject or absorb? This would depend on the power conditioning. Nearly
100% of capacity can be used if power conditioning is designed for this.
How long can it sustain this response? A disadvantage of supercapacitors is the short
sustain time, on the order of seconds to minutes.
Is there a particular ramp or shape to the response? There is a shape to capacitor response
so power conditioning is needed to form the response.

What determines each of these limits? Is there potential for it to be adapted? Hybrid
technologies, such as advanced batteries that combine conventional batteries with
supercapacitors, can ensure a very fast response that can be sustained for minutes to hours.
Where, when and how have these capabilities been demonstrated? The DOE Energy
Storage Database lists 28 projects with a total of 76 MW capacity that have been deployed,
mainly to provide transportation services and voltage support. They have also been deployed
in their use in advanced batteries.
In 2014, ENDESA in Spain commissioned a 4 MW (6 second) supercapacitor system to stabilize
voltage in La Palma in the Canary Islands. The system helps to avoid underfrequency load
shedding in a contingency event.
In Palmdale, CA, a 450kW supercapacitor from Maxwell Technologies provides 30 seconds of
storage as an uninterruptible power supply for a water treatment facility96.
At University of California in San Diego, a 28 kW supercapacitor system provides smoothing
and firming for a concentrated PV system97.
In early 2016, Duke Energy installed a hybrid energy storage system with a 277kW (8kWh for
a duration of nearly 2 minutes) supercapacitor98,99 and a 50kW (200kWh) battery that provides
fast solar smoothing and longer-term load shifting. Fast, high power responses are provided
by the supercapacitor to reduce thermal stress of the battery and not degrade battery lifetime.

95!SBC!Energy!Institute,!“Electricity!Storage”,!September!2013.!
96!Ibid.!
97!Torre,!W.,!Borneo,!D.,!and!Washom,!B.!“Energy!Storage!for!Integration!of!Renewable!Generation!at!the!University!of!California!–!
San!Diego”,!Electrical!Energy!Storage!Applications!and!Technologies,!San!Diego,!CA!Oct!20X23,!2013.!
98!https://news.dukeXenergy.com/releases/dukeXenergyXtoXputXnewXbatteryXandXsupercapacitorXsystemXtoXtheXtestXinXnXc!!
99!http://wininertia.es/projects/rankin/!!

80!
! !
!
What is the impact on the resource from responding in this manner? Provide a
quantitative estimate of any immediate (short run marginal) cost. There are virtually no
wear-and-tear costs for supercapacitors, so operating in this manner is not a problem. O&M
is very low. Cycle life is typically estimated at 100,000 cycles. One manufacturer estimates 1
million cycles with 20% reduction in rated capacitance100.
How mature is this technology? Is there potential for these characteristics to improve in
the future? To what degree, and over what timeframe? This technology is commercially
available and is based on the mature battery and capacitor industries. However,
supercapacitors have not been deployed much for utility-sector applications and therefore,
there is likely to be some potential for reduced costs and improvements in power conditioning
with widescale deployment.
What is the potential for false or spurious triggering? How is this managed? As with the
other inverter-based technologies in this section, this depends on the RoCoF detection scheme
(see section 2.2.2.2).
Are there any other important considerations for the use of this technology to provide a
fast frequency response service, to mitigate high RoCoF? For utility sector applications, this
should not be an issue but self-discharge rates are reported to be ~5-40%101. Supercapacitors
are adversely affected by high temperatures but this should be manageable in system design.
This does mean that overheating can result if an supercapacitor were exposed to a ripple
current, but RoCoF applications should not cause this issue.

3.6! Solar PV
Solar Photovoltaic (PV) generation is enjoying a worldwide explosion in growth. The two most
basic components of PV for provision of AC power are common to all applications, in power
ratings ranging from a fraction of a watt to utility scale projects approaching (or even
exceeding) a Gigawatt. The first component is a collection of photosensitive semiconductor
cells that when subjected to photons of appropriate wavelength, produce energetic electrons
that will flow, when allowed to do so: that is, they produce direct voltage and current. The
product of this DC voltage and current is power, which must be converted to 50Hz AC power
by inverters. PV installations consist of one or more of these modules, normally connected in
parallel. For projects of utility scale, AC collector systems (sometimes ‘reticulation’ in Australia)
resemble those of wind plant, with a supervisory control providing intelligent interface
between the grid and the farm. The most basic concept is shown in Figure 29.

100!http://www.maxwell.com/!!
101!Luo,!X.,!Wang,!J.,!Dooner,!M.,!and!Clarke,!J.!“Overview!of!current!development!in!electrical!energy!storage!technologies!and!
the!application!potential!in!power!system!operation,”!Applied0Energy!137!(2015)!511X536.!!

81!
! !
!
 Figure 29 Basic PV module102

3.6.1! Solar PV Components and FFR

For this discussion of possible fast frequency response, a few physical elements of PV are
critical:
1.! DC Power Rating. Individual photocells typically have voltage rating on the order of
2V, and current rating very roughly proportional to the area of the cell. The details of
voltage, rating, sensitive wavelength, temperature sensitivity, etc. vary with different
cell designs and materials, and are not particularly important here. The DC rating of
the collection of cells (i.e. a “panel”) dedicated to a specific inverter is important. The
DC rating (for this discussion) is the maximum DC power that can be produced under
conditions of maximum insolation (sunlight energy intensity at the panel). DC rating
is independent of the inverter.
2.! AC Power Rating. The inverter serves the function of converting the DC power to AC.
The cost of the inverter is dominated by the AC current rating, although the rating is
typically given in kVA at nominal AC voltage. The key point is that there is no

102!www.solarprofessional.com!

82!
! !
!
 fundamental requirement that rating of the inverter “match” the DC rating of the panel.
In practice, this has significant implications for FFR, as will be discussed below.
3.! Tracking. The orientation of the panel relative to the sun dictates how much of the
available insolation energy is converted. In general, PV installations can be fixed, single
axis trackers, or dual axis trackers. Tracking installations physically orient the panels
so that they capture more energy over the course of a daily solar transit. For a given
day and DC rating, the DC power available from the 3 types of tracking (for similar DC
rating) are conceptually shown in Figure!30.

Figure 30 Illustration of DC Power Impact of Tracking

4.! Time of Year. The impact of time of year is illustrated in Figure!31. The shorter days and
lower incidence angle affect the energy production. This seasonal variation has some
importance, relative to the inverter rating, as will be discussed below.

83!
! !
!
 Figure 31 Illustration of Time of Year Impact

Figure 32 Maximum Power Point Tracking Illustration

84!
! !
!
 5.! Maximum power point tracking (MPPT) control. The amount energetic electrons
liberated from the photovoltaic cell is sensitive to the voltage across the cell, the
amount of insolation and the temperature of the cell. In Figure 32 the relationship
between DC voltage, current and power is illustrated for one level of insolation and
cell temperature. The insolation and temperature of the cell vary continuously, with
insolation moving quite rapidly under some cloud conditions. One of the functions of
the inverter is to maintain the best DC voltage for the present condition, by means of
a maximum power point tracking (MPPT) control. The MPPT is the means by which the
solar PV maintains the minimum of: maximum available DC power, maximum AC
power, i.e. the inverter power limit; power setpoint, i.e. a curtailment. The MPPT is
normally a slower closed loop control than the inverter current control loop and the
supervisory voltage control loop. In order to maintain control stability between these
controls, their speed of response is typically separated by an order of magnitude.
Consequently, the MPPT control has a time constant on the order of a second.
5.! Supervisory control. For utility scale PV installations (i.e. those which are most likely to
be able to provide FFR), there is typically supervisory control that has many of the same
functions as a wind plant supervisory control. The supervisory control will accept
curtailment (and other) instructions from the host utility, and communicate them to the
individual inverters for implementation. As in wind plants, there are latencies in control
and communication paths that limit how fast the supervisory control can transfer
commands to the individual inverters. And again, like with the wind plants, the industry
has not yet been required to drive these latencies to very low levels.

3.6.2! Curtailment, Overload and Provision of FFR

Unlike wind, there is no inherent energy storage available via kinetics (i.e. the rotating mass).
Therefore, if a PV resource is to deliver any fast boost in power output, the inverter must be
operating at an AC power transfer level that is LOWER than the available DC power from the
PV panel at that point in time. Under conditions when the AC inverter (and balance of plant) is
not the limiting condition, this requires that the power production be curtailed to a level below
the available power. This is a nearly perfect analog to the pre-curtailment of the wind power
necessary to provide sustained FFR or PFR, as described above in the wind section. Note that
it is possible to have (battery or supercapacitor) energy storage integrated with the solar PV,
but aside from possible savings associated with shared infrastructure (inverters, transformers,
balance of plant), there is no intrinsic performance synergy associated with co-location (or
hybridization) of the energy storage with PV. Otherwise, the energy storage discussion in the
previous sections applies.

The steady-state and short term overload rating of the PV inverter can play an important role
here. While it is natural to think of the rating of the inverter in terms of kW, or kVA, the physical
reality is that the rating is dominated by the current carrying capability of the semi-conductor
valves. A simple illustration of the concept is shown in Figure!33. This figure is greatly simplified,
but captures the predominant character of the inverters not only for PV, but also for type 4
wind turbines, and for essentially all inverter-based energy storage devices. The figure is for
a four-quadrant device (i.e. power can come in from the grid on the left – obviously only
85!
! !
!
meaningful for energy storage and loads), and provide reactive power to the grid (northern
hemisphere) or consume it (southern hemisphere). A key point is that P (active power) and Q
(reactive power) are controlled independently as long as the device is not at its limit. But when
the combination of P and Q demand drives the operating point against the circle, the control
must decide which takes priority. The issue is further complicated by the reality that these
limits are really current limits: the figure as labeled is meaningful for nominal terminal voltage.
But when the voltage rises, the circles get bigger, and when it sags the circles shrink. (There
are several additional nuances here, but they are not very important for this discussion).
Finally, the outer circle, which represents overload capability, is complex. The degree of
overload that can be tolerated is a function of design and time. Depending on the inverter
design, this overload capability may be very small and very short…i.e. a few percent overload
for fractions of seconds. But other designs may have more capability, either as a result of
more robust design for other considerations, or as a deliberate incremental capability for
short-term overload.

Figure 33 Inverter Rating Concept

Overall, this trade-off between active and reactive current capability, and system voltage, can
be important for FFR. Two, not mutually exclusive, options exist:

1.! Give active power priority over reactive power. This especially makes sense if the
inverter is operating under-excited (i.e. consuming reactive power). It will raise the
voltage, allowing more power to be delivered for a given current. Conversely, when the
inverter is supplying reactive power, reducing the VARs to make room for active power
will decrease the voltage, increasing the current necessary to deliver the same active
power. It will be systemically dependent, and it is possible that the voltage reduction
will negate the benefit or create other systemic problems. These considerations are
situational, and will tend to be greater in weak grid situations.

86!
! !
!
 2.! Drive the inverter into the overload range. This must be done carefully, and with
knowledge of the design capability of the inverter. Considerations such as ambient
temperature and pre-disturbance valve junction temperature might be important, if
squeezing every last bit of power out is important.

These points only make sense if it is the inverter that is limiting the delivery of active power.
The inverter cannot cause the DC panel to make more power than the maximum allowed by
the instantaneous insolation.

The interplay between these two options, the system requirements and the available DC
power can be rather confusing. Phasor diagrams help illuminate the relationships. Figure!34
shows the current for a normal operating condition (blue dot) in which the current limits of the
inverter are not active. The blue phasor, shows active current (Ip) being injected to the grid at
a particular level of DC power (insolation), given by the vertical dotted blue line. The reactive
current, Iq, in this quadrant is being injected to the grid to support voltage (presumably in
response to required closed-loop voltage regulation). As long as the total current, Itotal, is within
the rating circle, the active and reactive injections are independently controllable.

Figure 34 PV Current Phasor for Normal Operation

As available DC power increases, the operation moves towards the steady-state current limit.
In Figure!35, the DC power is assumed to have increased to the green vertical dashed line, while

87!
! !
!
the reactive current requirement remains the same. This new operating point, the green dot,
represents operation at the steady-state current limits of the inverter. In the event that the
system reactive power demands (for Iq) and the available DC power (Ip at this voltage) together
exceed the current rating, then the inverter must establish which function takes priority. If
reactive support takes priority, which is typically the case up to the power factor required by
interconnection requirements, then active power will be limited. Alternatively, reactive current
can be reduced to make room for active current injection.

Figure 35 Normal Operation at Steady-state Current Limits

The concept of giving priority to active power for high levels of insolation is shown in Figure!36.
Here, an even higher level of insolation, given by the orange vertical dashed line, results in a
reduction of reactive current injection. In the limit, the inverter could go to unity power factor,
without exceeding the steady-state current rating of the inverter.

88!
! !
!
 Figure 36 Active Power given priority within Steady-state Rating

The concept of taking advantage of possible short-term overload rating of the inverter, as
illustrated by the dashed purple circle, is shown in Figure!37. The short-time overload rating
circle can be time, temperature and operating history dependent. The two concepts, giving P
priority and using short-term overload capability, can work together to provide control options
for achieving FFR.

89!
! !
!
 Figure 37 Operation with Short-time Overload of Inverter

These current-rating and DC power relationships raise an interesting new option for FFR, as
discussed in the next section.

3.6.3! Inverter and Panel Rating

Historically, the biggest cost component in photovoltaic systems was the semi-conductor
panel. Further, until recently, the overall cost of PV energy was considerably higher than from
competing generation resources. Consequently, the economics of PV design tended to strike
a balance between using the absolute least expensive components for the rest of the system
(i.e. the balance-of-plant), including the inverter, and making sure that every kW that the
panels might produce would be delivered to the user (meter). Therefore, inverters were
commonly rated to meet the maximum DC power of the panel. Further, the inverters were
assumed to operate at unity power-factor, thereby eliminating any extra current rating (and
cost) associated with delivery of reactive power.

Recently two important changes have developed: the cost of panels has dropped
precipitously, making the relative cost of inverters in the overall system much larger, and grid
authorities are demanding (through grid codes, etc.) that PV provide voltage support and have
reactive power capability.
90!
! !
!
These factors drive different design decisions that are important for this discussion of FFR
capability. Specifically, it is becoming common for the steady-state active power rating of the
inverters to be less than the maximum DC power of the panels to which they are matched.
The concept is illustrated in Figure!38.

Figure 38 DC power (different trackers) vs. AC Inverter Rating

The implication of this rating difference for provision of FFR is potentially profound. If the
under-rated inverter is designed to have a degree of short-term overload and/or the inverter
is given active power priority over reactive power delivery, then there is extra active power
available for delivery to the grid. The concept is illustrated in Figure!39.

91!
! !
!
 Figure 39 FFR capability for reduced rating PV inverters

This means that during any period in which the insolation exceeds the steady-state capability
of the inverter, there will be essentially zero opportunity cost to the PV plant to provide FFR up
to the short-time limit of the inverter and the instantaneous insolation. The owner of the PV
will incur capital costs to have this capability. Further, this is new ground. There is not, at this
writing, industry precedence for this approach.

3.6.4! PV Experience with Fast Control Response

Recently NREL conducted extensive testing on a 20MW utility-scale PV plant in Puerto Rico103.
The plant, which was developed by FirstSolar and is owned by AES, was subjected to a number
of tests (as quoted directly from the NREL report):

”This demonstration showed how active power controls can leverage PV's value from
being simply a variable energy resource to providing additional ancillary services that
range from spinning reserves, load following, ramping, frequency response, variability
smoothing and frequency regulation, to power quality. Specifically, the tests conducted
included variability smoothing through automatic generation control, frequency
regulation for fast response and droop response, and power quality.”
Results for the fast frequency response tests are shown in Figure 40. Again, to quote the NREL
report:

103 www.nrel.gov/pv/news/2016/21630.html!!

92!
! !
!
 ”There was a delay of approximately 50-100 milliseconds between the frequency
change and the beginning of the FFR by the plant. Implementing sophisticated control
theory and fine tuning the power and current gains could help with minimizing the
oscillations shown in the traces below. In general, more care is needed to fine tune and
set controls to achieve better long-term performance. But with respect to the speed
and power increase of the frequency response, the tests were successful.“

Figure 40 FFR Tests for 20MW PV Plant104

Limits on speed of response. The response here (Figure 40) appears is on a similar scale of
speed to the BESS results presented above. The physics of the converter are nearly identical
between the two, but the character of the DC link is rather different and can have an impact
on the speed and amplitude of the response. Specifically, the DC voltage on the terminals of
a battery string is relatively “stiff” with a significant internal resistance. The resistance creates
a drop in the DC voltage when the discharge current jumps up (and conversely a rise, during
a jump up in charging current). Consideration of this droop is included in the current control,
which adjusts pulse width modulation (PWM) firing accordingly. The DC voltage for a PV panel
is less stiff, and itself relies on the response of the photosensitive semi-conductors. The net
result is that manipulation of the active current drawn from the panel has a substantive
impact on the DC voltage, which must be managed by the converter. This has the potential
to impact the speed of response. To-date, PV has not been required in commercial operations
to adjust active power in response to external instructions at extremely high speeds.
The fast response using pre-curtailment of the PV system overrides the MPPT control and
operates off of the PV module’s maximum power point. The inverters operate on the over-
voltage side of the MPPT curve shown in Figure 32, and when FFR action is needed, the MPPT
nearly instantaneously shifts the voltage to the maximum power point, providing an increase

104!Advanced!GridXFriendly!Controls!Demonstration!Project!for!UtilityXScale!PV!Power!PlantsU!V.!Gevorgian,!B!O’Neill,!NREL/TPX
5D00X65368,!January!2016.!

93!
! !
!
in power output. The DC voltage dips along with the response, and implications on system
stability need to be considered. The response time of this approach depends on the equivalent
input capacitance of the PV, which is small. However, recent tests on present full sized
(300MW) state-of-the-art utility scale PV projects suggest that response times of 100ms
should be doable. Further, PV inverter experts interviewed say it should be possible to do this
within 50ms of receiving an instruction.
Control Philosophy. The relative benefits and disadvantages of open-loop vs. closed-loop
FFR controls that were described in detail in the wind section apply to solar PV. However, this
discussion and debate has, for practical purposes, not begun in the industry.

3.6.5! Costs
There is not yet sufficient commercial experience to determine capital costs for FFR (or PFR) in
PV plants. Because FFR affects inverter software controls to a greater extent than physical
hardware (or inverter rating, which would impact copper and silicon content), the incremental
cost is thought to be modest. Historically, inverter costs have been a small fraction of the
overall PV plant cost. But, per the discussion in Section 3.6.3, this relationship in component
costs is changing. NREL’s Q1 2015 benchmarks for breakdowns of PV plant costs find that
inverters cost $0.29/W out of a total installed cost of $3.09/W for residential, rooftop PV, or
about 9% of the total cost105. If, hypothetically, PFR were to increase inverter costs by 25%,
the total rooftop PV system cost would increase by 2%. For utility-scale PV systems, inverters
cost $0.11/W out of a total installed cost of $1.77/W, or about 6% of the total cost106. A
hypothetical 25% increase in inverter costs would increase the total utility-scale PV system
cost by less than 2%. Again, as was noted with the wind discussion, a 1% to 2% increase in
capital cost can represent a substantial penalty against the margins or return on investment
for a project. This is especially true if the increased capital costs are not factored in while the
project is being designed and financed. Retroactive requirements are likely to be much more
expensive.

3.6.6! Synopsis of Solar PV FFR

How fast can it respond? Response times on the order of 100-200ms have been
demonstrated.
How much can it inject or absorb?107. Up to the short-time overload limit or maximum
insolation.

105!Chung,!D.,!Davidson,!C.,!Fu,!R.,!Ardani,!K.,!and!Margolis,!R.!“U.S.!Photovoltaic!Prices!and!Cost!Breakdowns:!Q1!2015!
Benchmarks!for!Residential,!Commercial,!and!UtilityXScale!Systems,”!NREL/TPX6A20X64746,!Sept.!2015.!
http://www.nrel.gov/docs/fy15osti/64746.pdf!!
106!Ibid.!
107!HDR!study!for!PacifiCorp,!July!2014!

94!
! !
!
How long can it sustain this response?. Up to the short-time overload time limit, if converter
limited, or sustained longer at the maximum insolation if the inverter is within its continuous
rating.
Is there a particular ramp or shape to the response? Trapezoidal behavior should be
possible. Closed-loop control response should be possible.

What determines each of these limits? Is there potential for it to be adapted?. There are no
torsional or mechanical limits; speed is primarily a function of control limitations, including
coordination with MPPT.

Where, when and how have these capabilities been demonstrated? Very limited. The NREL
tests quoted here were performed on a commercial installation. Several large PV plants in the
US are equipped with a full spectrum of voltage/reactive power controls, and accept
curtailment orders. We are unaware of any commercial installations that presently offer
frequency control.
What is the impact on the resource from responding in this manner? Provide a
quantitative estimate of any immediate (short run marginal) cost. For overload, as
described above, there should be no marginal cost. For curtailment, the opportunity cost is
that of the solar energy (and lost tax benefits/subsidies, if any).
How mature is this technology? Is there potential for these characteristics to improve in
the future? To what degree, and over what timeframe? Not mature now. Per detailed
discussion above, potential for development seems highly promising.
How is the ability to deliver a fast frequency response service affected by nearby power
system disturbances, such as transmission faults? Will the FFR service still be available in
the period following a transmission fault? How could this be managed? PV tolerance to
faults and high RoCoF can be good, but practice in some places has caused distributed PV to
have poor grid fault and frequency tolerance. The consideration of DC bus stiffness (as
discussed above) can be a contributing factor in the recovery following a fault. Maybe PV
systems completely block (no current) during voltage depressions and faults.

3.7! Load Based Resources

Load resources can provide high quality FFR. For an under-frequency, negative RoCoF event,
instead of asking a generator to rapidly increase power, we may elicit a faster response by
asking load to drop. For example, in ERCOT, loads provide half of the interconnected power
system’s Frequency Responsive Reserves that support the system in a sudden loss of
generation event. ERCOT’s FFR triggers at 59.7 Hz or 59.8 Hz and provides full response in 500
ms. Loads that provide fast response have under-frequency relays that drop the load when a
preset threshold is exceeded. Because loads, storage and switchable devices (e.g. dynamic
brakes) can provide FFR, it’s important to ensure that all resources are allowed to participate
in potential markets for these services. These resources may prove to be more effective and
economic than PFR provided by generation resources in maintaining system security.

95!
! !
!
For AEMO, participation by loads in an SPS that detects separation and other conditions which
warrant controlled load shedding (e.g. import overload that risks separation) makes a great
deal of sense. This option is integral to the extended analytical illustration provided below in
Section 6.

3.7.1! Inertia Considerations

One of the considerations in using load as an FFR resource is that load may also be providing
inertia, which is helping to mitigate RoCoF. Therefore, it may be necessary to assess the
characteristics of potential load-based resources and accept those that can provide the
greatest net benefit to the system. The power industry is struggling with development of good
data and understanding of the inertial contribution of loads for large grid events. As has been
noted before, the extreme nature of the problem facing South Australia makes this question
nearly uncharted waters.

3.7.2! Tripping vs. Blocking vs. Continuous Control

Modern commercial and industrial loads often have a substantial component of electronically
controlled motors and processes. This gives rise to a class of load participation in providing
frequency response that is more sophisticated and more valuable than traditional breaker
switched load shedding. As with the converters that control most of the other technologies
discussed here, the front-end converters for many loads can be temporarily blocked or
reduced in power levels in response to an appropriate signal. That signal could be locally
measured frequency or a trigger (as from an SPS). The economics of using load this way varies
a great deal. Considerations include the cost to the provider of interrupted or reduced
throughput, capital costs, restart costs, to name a few. Economies of scale very much apply
here. In the US, one of the more active participants in providing this service are aluminum
smelting “pot lines”, which can interrupt power consumption very rapidly and with little
economic consequence if the duration is kept within bounds. Other loads, such as heating
and cooling, tend to be good candidates.

3.7.3! Highly Distributed Responsive Loads

There is a practical limit to how much industrial load may be available to participate in fast
voluntary load shedding. Therefore, it may be advantageous to consider means to get FFR
from highly distributed loads. Rapid communication to a larger number of load nodes appears
to be gaining practicality, as the Arizona example cited in Section 2.2.2.1 describes.

Another option maybe autonomous load controls. For example, in the UK, the National Grid
Company (NGC) has introduced a special tariff which rewards large-scale consumers who
agree to provide a limited form of frequency response. Called “Frequency Response by
Demand Management” (FCDM), the scheme involves the placing of certain large loads behind

96!
! !
!
frequency-sensitive relays which isolate the load when the frequency falls below a pre-set
level, often 49.7 Hz108. Pacific Northwest National Laboratory in the US is promoting "Grid
Friendly Appliances" which are similar to dynamic demand appliances being promoted in the
UK. The use of highly distributed loads to provide FFR is likely to be subject to the constraints
and uncertainties of rapidly measuring RoCoF and frequency as enumerated above. Waiting
until frequency drops by (say) 300mHz in South Australia uses up a significant amount of the
available time for FFR. Information on how quickly small domestic devices (e.g. water heaters,
pool pumps) might respond to stop/block signals is limited, but the underlying physics
suggests that rapid current interruption should be possible. Hydro Quebec has a pilot project
for frequency sensitive control of hot water heaters. They aim to trigger within 250ms. For
very fast actuation, again the pressing issue is likely to be mis-triggering associated with
having frequency or RoCoF triggers too tight. Many classes of loads will tolerate occasional
interruption, but will be unwilling to tolerate (for example) multiple interruptions per day.

To get higher fidelity trip signals, variations on transfer trip are required. The types of
communication techniques discussed in Section 2.2.2.1 (and installed at Salt River Project,
Arizona in the US) are suitable for rapid communication of trip signals to a large population of
activating nodes. The challenge is likely to be identification of enough activating loads, and
justification of the well-developed fiber-optic communication necessary to get the speed of
response required in the South Australia situation.

One of the significant challenges associated with tripping highly distributed load is to know
how much load will be disconnected at any given time. The activation of load blocking is, by
its nature, discrete. The load is either blocked or not. Thus, droop or proportional response
requires sophisticated distribution and time staggering across many binary triggering signals.
If this embedded UFLS response represents a modest portion of the total response, tuning or
differentiation is less important. But when this approach represents the majority of the
response, then direct measurements, state-estimators, or other schemes to determine the
amount of load that will be tripped at any present instant are needed. The technology exists,
but the data gathering and computational burden for this approach are not trivial.

3.7.4! Parallels to Industrial Facilities

The situation faced by South Australia has strong parallels to problems occasionally faced by
large industrial facilities that generate some of their own electricity, and purchase some from
the host utility grid. GE has experience in design and evaluation of schemes for systems that
allow “trip to island”. As with South Australia, in situations where opening the connection from
the facility to the grid results in excessive power-load unbalance, tripping of some loads is
required in order to maintain stability. It is common for industrial facilities (e.g. chemical,
refineries, manufacturing,..) to have a range of sensitivity for their loads, from highly sensitive
(i.e. must not trip due to safety or high costs) to low sensitivity (i.e. easily tolerates occasional
interruption). Differentiation between these loads makes it difficult to send “simple” breaker
open signals, since that can result in critical loads being interrupted. Under such

108!http://www.dynamicdemand.co.uk/current_work.htm!

97!
! !
!
circumstances, it is possible to use (a) signals passed through motor-control centers, that can
differentiate the loads, and shut-down or back down those that can be tolerated, or (b) to
place RoCoF or UFLS relays directly on the interruptible loads. Again, the primary question is
one of speed for either option, and a further question for the local relay approach is the
consequence of false triggers.

3.7.5! Synopsis of Load Based FFR

How fast can it respond? Communication to substations (TSignal) is possible within 2-3 cycles.
Blocking (TActivate+9TActivateFully) depends on the load to be interrupted.

How much can it inject or absorb? This will be dependent on the sophistication of the state
estimator (I.e., the ability to continuously and accurately track the amount of load that can be
tripped by the FFR scheme at any given time).
How long can it sustain this response? Depends on the blocking mechanism, but assuming
that relief is sustained until released by the grid operator is a good default.
Is there a particular ramp or shape to the response? Response is step-wise.
What determines each of these limits? Is there potential for it to be adapted? Speed of
detection; speed of blocking.
Where, when and how have these capabilities been demonstrated?. Examples of fast load
tripping have been implemented at Salt River Project, Arizona and Southern California Edison
in the USA. See Section 2.2.2.1.
What is the impact on the resource from responding in this manner? Provide a
quantitative estimate of any immediate (short run marginal) cost. Customers will estimate
opportunity and lost-revenue costs, based on specifics of load interrupted.
How mature is this technology? Is there potential for these characteristics to improve in
the future? To what degree, and over what timeframe?. This is well established art, in slower
time frames. The primary issue and potential for advancement (and application to South
Australia) is whether time delays and latencies can be significantly reduced, without creating
unacceptable risk/frequency of mis-operation. With the right economic signal, this
technology could advance substantially.
How is the ability to deliver a fast frequency response service affected by nearby power
system disturbances, such as transmission faults? Will the FFR service still be available in
the period following a transmission fault? How could this be managed? This is primarily an
issue of mis-operation, which depends on how and where the trip command is generated. If
the trigger is directly initiated by the opening of the AC intertie, detection is not sensitive to
post-fault distortions in voltage/current waveforms. And provided the trigger signal can be
quickly and securely transmitted to the loads being tripped, FFR action should be very
dependable. However, if the trigger signal is determined locally (near the load) by a frequency
or RoCoF transducer, then there is some risk that distorted waveforms or local dynamic effects

98!
! !
!
during and immediately following a fault could cause mis-operations (i.e., falsely triggering
when not needed or failure to trigger when needed).

3.8! HVDC Transmission

This discussion focuses on the ability of HVDC to contribute to FFR during the period before
the frequency nadir (i.e., the frequency arresting capability of HVDC).

3.8.1! Physics
HVDC power transfer is regulated by power electronics and programmable controls which are
able to respond in a small fraction of a second. This capability can be used in a few different
ways to respond to a power-load imbalance (either underfrequency or overfrequency) in an
islanded system.

Fast ramp down in HVDC power (runback). This is a commonly applied control feature where
power is rapidly reduced in response to an overfrequency event. For Voltage Source Converter
(VSC) systems, AC bus voltage is regulated by the HVDC control system. For conventional line-
commutated converter (LCC) systems, power reductions can produce AC overvoltages if there
is an excess of reactive compensation, so some reactive power banks may need to be tripped.
This is mature technology.
Fast ramp up in HVDC power. This control feature is less commonly applied and it is
significantly more challenging than runbacks. The speed and performance of a ramp-up
scheme depends on the strength of the AC system. That is, the AC grid must be able to support
higher power flow from the HVDC terminal without significant depression of the AC bus
voltage. For VSC systems, the HVDC converters can contribute reactive power support within
the ratings of the converters. For LCC systems, increased HVDC power results in increased
reactive power consumption by the converters, which can cause significant depression of the
AC voltage unless additional reactive power can be supplied from other devices (e.g. SVC).

Fast ramp up with short-term HVDC overload. This scheme would temporarily increase the
HVDC power transfer above the normal system rating and then return to rated power transfer
after a few seconds. This type of scheme could be helpful in arresting the frequency decline
during an underfrequency event, during the critical period when generator governors are
beginning to respond but have not yet achieved their full response.
Normally, the maximum power transfer of a HVDC system depends on the ratings and
capabilities of the various components within the system. For example, converter
transformers can withstand significant short term overloads due to their large thermal mass.
Converter valves, on the other hand, are constrained by the maximum allowable junction
temperature within the power-electronic devices (e.g., thyristors). These devices have very low
thermal inertia due to their small size, and junction temperatures increase very rapidly (within
a few seconds) when current loading is significantly increased. Such temperature changes
are too fast and too localized to be influenced by the valve cooling system.

99!
! !
!
There is very little experience with short-term (several seconds) overload of HVDC systems.
However, one recent example is the Champa-Kurushetra HVDC system109 in India. The valve
design included the following overload capabilities; 31% above rated for 2 hours, 65% above
rated for 5 seconds. Of interest, here is the 5-second overload capability which could be
beneficial for a FFR function.
All converter valves have some inherent level of overload capability. But if such capability was
not specifically required in the specification of the system, the valve’s capability is probably
unknown. However, it should be possible for the manufacturer to review the design of an
existing valve to determine the potential for short-term overload operation. However, this
would only be the first step. If overload capability is determined to be feasible, the
manufacturer and owner would need to work together to develop a workable scheme to
enable the short-term overload functionality while maintaining the reliability and security of
the HVDC valve equipment.

Another approach for adding overload capability to an existing HVDC system would be to add
another valve group in parallel to the existing valves, thereby increasing the total current
rating of the HVDC terminal. The rating of the parallel valves would be coordinated with the
short-term overload capability of the HVDC cable. The Pacific HVDC Intertie in the USA was
originally built with 2000 ampere valves, and later augmented with 1100 ampere valves in
parallel. This approach would be very expensive, and other alternatives such as battery
storage systems would likely be more cost-effective.
Speed of Response
Several factors affect the speed of response for rapid changes in HVDC power, especially for
increases in HVDC power.
Inherent time constants in the HVDC system itself. Control systems can respond quickly, but
the rate of change of current can be limited by the electrical characteristics of long HVDC
cables. In a LCC-type HVDC system like BassLink, a severe AC fault in the vicinity of a converter
terminal can cause the DC line or cable to discharge, and the DC-side voltage may approach
zero. When the AC voltage recovers, the voltage on the DC line or cable must also be restored
to near nominal magnitude before the HVDC system can transmit significant power. For the
HVDC system to recover, the converters must first inject current into the line/cable to “charge
up” the line/cable capacitance and restore DC-side voltage. After that occurs, the inverter
terminal can begin injecting current to the ac system, thereby restoring HVDC system power
transfer. The time required to charge up the line/cable capacitance, restore voltage, and
restore power transfer depends on the system-specific electrical characteristics of the HVDC
and AC systems. Commissioning tests of BassLink demonstrated a total recovery response

109!Thyristor!valve!for!the!12Xpulse!converter!for!the!ChampaXKurukshetra!HVDC!transmission!scheme,!!Mohammad!Hassan!
JodeyriU!Andrzej!Dzus,!!2013!IEEE!Innovative!Smart!Grid!TechnologiesXAsia!(ISGT!Asia),!!Year:!2013,!!Pages:!1!X!6,!DOI:!
10.1109/ISGTXAsia.2013.6698790!

100!
! !
!
time near 300 ms for a DC-side fault.110 See Figure 41, which is an excerpt from the referenced
publication. Although this plot is for a DC-side fault, an AC fault that discharges the DC-side
line/cable would exhibit a similar recovery response. And although this response is slow in
comparison to other HVDC projects, it is potentially fast enough for a frequency-arresting
function.
AC system voltage support. HVDC power transfer depends on the ability of the ac system to
maintain the bus voltage at the converter terminal nearly constant. If HVDC power transfer is
increased at a rate that exceeds the capability of the AC system voltage support, the voltage
will sag and system security will be at risk. System studies are needed to quantify the ability
of the ac system to support rapid increases in HVDC power.
Type of control scheme. Fast changes in HVDC power can be initiated by either an open-loop
triggered action scheme (e.g., detect breaker opening and initiate predetermined action) or
measured-frequency input scheme (a frequency governor or modulation function). (See
additional discussion below.)

Source: Latest control and protection innovations applied to the Basslink HVDC interconnector, M. Davies et al, 2006

Figure 41 Response of BassLink to DC-Side Fault

3.8.2! Controls
There are several options for implementing frequency arresting controls on HVDC systems:
1.! Frequency-sensitive power modulation control function that changes the power order
of the HVDC system as a function of grid frequency. This system typically includes a
frequency transducer (with voltage and/or current inputs), deadband, gain/phase
compensation, and output limits.
2.! Triggered action run-back or run-up scheme that changes the power order of the
HVDC system by a preset amount in response to a given event (e.g., trip of a specific

110!Latest!control!and!protection!innovations!applied!to!the!Basslink!HVDC!interconnector,!!M.!DaviesU!A.!KolzU!M.!KuhnU!D.!
MonkhouseU!J.!Strauss,!!AC!and!DC!Power!Transmission,!2006.!ACDC!2006.!The!8th!IEE!International!Conference!on,!!Year:!
2006,!!Pages:!30!X!35,!DOI:!10.1049/cp:20060007!

101!
! !
!
 generator or transmission line). Such systems often have multiple triggering events
with different power order responses.
3.! Combination of both triggered action and frequency sensitive power modulation.
The frequency modulation approach has the benefit of being continuously active, and the
magnitude of the response is driven by the magnitude of the disturbance and the grid’s
frequency excursion. To achieve fast response, these types of controls require high gain and
short time constants, which can unfortunately cause undamped oscillations or instabilities
under some system operating conditions. The Pacific HVDC Intertie111 and the Intermountain
HVDC system112 both experienced unstable oscillatory grid events due to such controls when
gains were set too high with the intent of achieving fast response. Figure 42 shows one such
event in the western USA grid involving the Pacific HVDC Intertie.

Source: Robust damping controls for large power systems, J. F. Hauer, IEEE Control Systems Magazine, 1989

Figure 42 Unstable Grid Event Involving HVDC Frequency-Input Modulation Controller

The triggered action approach initiates an open-loop pre-programmed step change in HVDC
power that is pre-calculated to rapidly adjust the grid power balance to more stable condition.
It has the benefit of speed (can change HVDC power in a fraction of a second) without the risk
of control-loop instabilities inherent with frequency modulation controls. Its response is
limited to pre-determined actions for specific triggering events and specific operating

111!Robust!damping!controls!for!large!power!systems,!J.!F.!Hauer,!!IEEE!Control!Systems!Magazine,!!Year:!1989,!Volume:!9,!
Issue:!1,!!Pages:!12!X!18,!DOI:!10.1109/37.16744.!
112!1994!System!Disturbances!XX!Review!of!Selected!System!Disturbances!in!North!America,!North!American!Electric!Reliability!
Council,!October!1995.!

102!
! !
!
conditions (e.g. flow on an AC interconnector). Individual triggers can be armed when needed
and disarmed when not needed.

The combination approach uses a triggered action to rapidly change HVDC power to an
appropriate post-disturbance level with a slower-acting frequency modulation control to fine
tune the post-disturbance operating point. This is a promising approach when high-risk grid
events are well known and directly observable (i.e., trip of a line or generator).
New Zealand, Pacific HVDC Intertie, Sardinia-Corsica-Italy HVDC, Intermountain HVDC,
Gotland HVDC, Square Butte HVDC, and many other systems have power modulation
controllers that respond to changes in grid frequency. In parallel AC/DC systems, they provide
damping to inter-area oscillations and in some cases, improvement in first-swing stability
following disturbances. For isolated or island systems, the HVDC provides governing response
like a generating plant – i.e., contribute to frequency regulation using droop control. In these
systems, the controllers were custom-designed using dynamic simulation tools to design
system specific settings to balance dynamic response and stability over a wide range of
system operating conditions.

3.8.3! Possibilities and Trade-Offs with HVDC for Fast Frequency Response
The preceding discussion related to the physics and experience with HVDC systems
performing functions that could contribute to fast frequency response. The following
discussion is intended to provide some overall perspective with respect to the challenges
faced by the AEMO grid.
The most critical events leading to rapid frequency decline is for loss of the Heywood
Interconnector) which leaves South Australia as an electrical island with a generation
shortage. This event is most severe when flow on the Heywood interconnector is at or near
maximum power flow from Victoria to South Australia before the event. When Heywood is
heavily loaded, MurrayLink is likely to be loaded at or near its maximum as well. This limits the
possibilities for MurrayLink loading to be transiently increased to help mitigate the South
Australia unbalance.
If frequency decline is rapid, then the response time needs to be as fast as possible. The total
response time would include the time to detect the event, the time to communicate that
information to the HVDC terminal, and the time for the HVDC system to respond.
If the event is not initiated by an ac fault (e.g., if the ac tie simply trips or if a South Australian
generator trips due to an auxiliary system problem), then the response time for a step change
in HVDC power could be very fast, perhaps within 50 to 100 ms. However, if the event is
initiated by an AC system fault, then there is the possibility that the HVDC system voltage might
be affected and that the DC-side line voltage would need to recover before HVDC power can
be moved to a new (higher) level. HVDC system response time would be much longer that 100
ms (perhaps near 300 ms).

103!
! !
!
Given that the HVDC response (due to physics) may be slower than desired, it is therefore
critical to reduce the event detection and communication times as much as possible in order
to minimize the total response time. This is commonly achieved in remedial action schemes
using the following approach:

•! Detection: Monitor the opening of critical breakers using “B” contacts on the breakers
themselves, or breaker status information in protective relays. Detection time is less
than one cycle from breaker opening.
•! Communication: Send breaker status signal by dedicated communication channel to
the HVDC terminal, using communication systems that interconnect the transmission
substations. Similar remote status signals are commonly implemented as part of
transmission protection schemes, and have used power-line carrier, microwave, or
fiberoptic communication channels.
Some systems have implemented frequency-input governor type schemes with the intent of
eliminating the need for communications. However, frequency input systems introduce
additional challenges that affect the total speed of response.
Frequency transducers use AC bus voltages as inputs, and the voltage waveforms become
very distorted during AC system faults and during the recovery period after the fault is cleared.
To keep the transducer from producing erroneous outputs, it is necessary to slow down or
filter the transducer output signal.

If fast response is needed from a frequency input controller, then the controller gain must be
high and the time constant must be short. But excessively high gains and/or short time
constants reduce the closed-loop stability margins to a point where widespread spontaneous
or growing oscillations may occur on the power grid. To avoid such oscillations, the gain must
be limited, which also limits the speed of response.
The bottom line is that there is no obvious “best choice”. A reasonable path forward is to
gather the necessary information to quantify each of the factors discussed above and to then
select the most promising approach. The first step would be to initiate a study that models
the performance of the MurrayLink HVDC system (using EMTP or PSCAD type models),
quantifies their response characteristics, and determines their capability to perform additional
functions to assist with fast frequency control.
How fast can it respond? Speed of response depends on two factors that are unique to each
application.

•! The design of the HVDC system. Back-to-back HVDC systems with strong AC systems
can respond in 50 ms. Long HVDC cables connected to weak AC systems have
response times in the range of 200 ms to 500 ms.
•! The type of initiating event. If the grid event does not involve an AC fault near the
HVDC terminal, HVDC response time is very fast, in the range of 50 to 100 ms. If the
grid event includes an AC system fault that discharges the HVDC line or cable, then the

104!
! !
!
 response time includes the fault recovery time of the HVDC system, which is in the
range of 100 ms to 500 ms, depending on the strength of the AC system.
How much can it inject or absorb? Any power level within the rating of the HVDC converter
equipment, including any short-time overload capability.
How long can it sustain this response? Indefinitely, for any power level within the HVDC
converter ratings. Short term overloads are limited by the thermal time constants of the
semiconductor devices, which are typically in the range of 2-5 seconds.
Is there a particular ramp or shape to the response? No. But the rate of response may need
to be constrained if the AC system has insufficient dynamic voltage support to prevent
collapse of the ac bus voltage when the HVDC power is rapidly increased.
What determines each of these limits? Is there potential for it to be adapted? Lots of
factors determine the limits. Response is highly system-dependent per the preceding
discussion. Detailed EMTP-type simulation studies are required to determine actual
performance capabilities for the MurrayLink HVDC system.
Where, when and how have these capabilities been demonstrated? HVDC power runback
schemes are very common and are offered as standard feature on most HVDC systems. New
Zealand, Pacific HVDC Intertie, Sardinia-Corsica-Italy HVDC, Intermountain HVDC, Gotland
HVDC, Square Butte HVDC, and many other systems have power modulation controllers that
respond to changes in locally measured grid frequency. The authors are not aware of any
specific systems that include HVDC power run-up (step increase) functions, but the
implementation would be the same as for runback. However, schemes to rapidly increase
power require more technical analysis to quantify performance capabilities and limitations.
What is the impact on the resource from responding in this manner? Provide a
quantitative estimate of any immediate (short run marginal) cost. No impact on HVDC
converter equipment, provided the control actions are designed to keep currents and voltages
within the rated capabilities of the equipment.
How mature is this technology? Is there potential for these characteristics to improve in
the future? To what degree, and over what timeframe? Both triggered response functions
(runback/run-up) and frequency modulation functions are mature technology. Use of HVDC
short-term (2 to 5-sec) overload is less mature, as very few systems have included such
capabilities in their design specifications. The Champa-Kurushetra HVDC system is the only
example found during the search of published literature.
How is the ability to deliver a fast frequency response service affected by nearby power
system disturbances, such as transmission faults? Will the FFR service still be available in
the period following a transmission fault? How could this be managed? Nearby AC
transmission system faults have significant impact on:

•! The speed of response of the HVDC terminal, since the HVDC must recover to prefault
power flow before it can provide additional power for FFR.

105!
! !
!
 •! The ability to rapidly and accurately transduce frequency after the fault, due to voltage
distortion caused by the ac fault and by the subsequent recovery of the HVDC
converters.
In summary, the preponderance of industry experience suggests that the most effective and
secure approach for HVDC to contribute to FFR is a remedial action scheme (RAS) that detects
the most critical events via breaker status, transmits a triggering signal to the HVDC terminal,
and initiates a pre-programmed action designed for the specific event. With this approach,
event detection and signal transmission happens in parallel with the ac and HVDC system
fault recovery, and by the time the HVDC converter is recovered and able to respond, the
communicated signal arrives and triggers a specific open-loop action. This scheme enables
a large response in a very short time, and avoids the significant risk of widespread oscillatory
instabilities associated with frequency-input HVDC power modulation schemes.

106!
! !
!
4! IMPLEMENTATION RISKS AND CONSIDERATIONS
This section discusses a range of practical considerations which may be common to a solution
of the frequency control problem.

4.1! Torsional Impacts on Turbine-Generators

Very fast active power injection by any inverter-based FFR resource will cause torsional
impacts on synchronous machines operating in the electrical vicinity in the power system.
Amplitude and rate of rise of power injection, coupled with electrical proximity to operating
synchronous machines dictate the degree of impact. Due diligence requires that potential
risks, including potential loss of life, especially for steam units, be considered. These concerns
apply, generally to a lesser extent, to wind generation, particularly type-1 and type-2
machines. Transient torques from open-loop step inputs, and closed-loop stimulation of
torsional oscillatory modes are both risks. Both of these concerns are particularly acute if
there is risk of frequent inadvertent triggering of response.
For systems using frequency-input power modulation, there is a risk that the controller may
reduce the inherent damping of torsional vibrations to a point where spontaneous growing
torsional oscillations113 may occur. [This was the case for the Square Butte HVDC transmission
system in the USA114. It was necessary to modify the control parameters and include torsional
notch filters to reduce the interaction to safe levels]. The design of frequency-input
modulation controllers should investigate the risk of torsional interaction in advance, and if
torsional interaction risk is found to exist, mitigation and protection features must be included
in the design.

4.2! Fault induced voltage and power depression

Grid events involving large frequency excursions are often triggered by faults in the AC
transmission system. Aside from the temporary interruption of AC power flow, the faults
produce a few other detrimental consequences that affect FFR:

•! Operation of power electronic devices in the vicinity may be seriously disrupted. HVDC
converters as well as converters for FFR devices may experience power flow
disruptions to low or distorted voltages, both during the fault and for some period after
the fault as the converter control systems resynchronize to the post-fault system
voltage.
•! Transducers measuring frequency or RoCoF are more likely be unable to accurately
determine grid frequency during faults and for a short period after fault clearing due

113!The!turbineXgenerator!driveXtrain!of!synchronous!thermal!generation!and!wind!turbines!have!multiple!masses!–!at!least!the!
turbine!and!the!generator!X!connected!by!a!shaft.!!Oscillations!of!this!massXspring!system!put!stress!on!the!shaft.!
114!M.!Bahrman,!E.V.!Larsen,!R.J.!Piwko,!H.S.!Patel,!"Experience!with!HVDC!X!TurbineXGenerator!Torsional!Interaction!at!Square!
Butte",!IEEE!Transactions,!PAS,!Vol.!99,!MayXJune!1980,!pp.!966X975.!

107!
! !
!
 to the distorted voltage waveforms that exist during the fault and during the
subsequent recovery of the power electronic converters.
Quantifying these effects would probably require a system study using EMTP-type models to
simulate the fault events and determine the voltage waveforms that occur during and after
the fault events. To do this right would likely require a massive EMTP study of the South
Australia region.
As noted in Section 3.8, the vulnerability of the HVDC to the voltage depression associated
with a fault that trips the AC tie line puts the power from the DC line at risk just as the system
needs it most. That is, HVDC power may be interrupted or reduced during the fault and it
would need to recover before providing any contribution to FFR. Of the various technologies
discussed in Chapter 3, the large (non-inverter type) flywheel technologies have particularly
good characteristics for reducing this risk. Colocation of these devices in the electrical
proximity of the HVDC terminals could have benefits for the fault recovery performance of the
DC.

Wind turbines, PV and other devices may well have blocked power flow during significant
voltage depressions. Physics limits the amount of power that can be injected into a low
voltage. That makes the overall FFR issue more complex as well. For all the subject
technologies, guaranteeing fast response in light of this complex behavior adds to the
challenge. Further, the ability of the various inverter-based FFR technologies to start (or
resume) power injection immediately after a fault will vary with technology and equipment
design. In general, as mentioned above, devices that are more resistant to change in the DC
link voltage of the inverter, such as batteries, will tend to be able to inject power sooner
following AC voltage recovery.
For additional context, EirGrid and SONI are implementing a new system service product to
quickly recover MW output following a disturbance: Fast Post-Fault Active Power Recovery
service. Following any fault that is cleared within 900 ms, including transmission faults, a plant
providing this service must recover its active power to at least 90% of its pre-fault MW output
within 250 ms of when the voltage recovers to at least 90% of its pre-fault level115. This service
is a market mechanism to encourage faster recovery than the minimum required by the grid
code. It was created in recognition of the difficulty of meeting this requirement, and is also an
incentive for existing (e.g. grandfathered) resources to invest in speeding up their
performance.
EirGrid/SONI’s phase 2 RoCoF study finds that for Ireland to maintain RoCoF within 0.5 Hz/s,
they need synthetic inertia to begin responding by 100ms from the start of the event (as visible
to the device) and to achieve full response within 200 ms from when the response begins116.
This has not been translated into a service yet.

115!EirGrid!and!SONI,!DS30System0Services:0Portfolio0Capability0Analysis,!November!2014.!
116!EirGrid!and!SONI,!RoCoF0Alternative0&0Complementary0Solutions0Project0Phase020Study0Report,!March!31,!2016.!

108!
! !
!
4.3! Voltage control and collapse (weak AC system)
Rapidly increasing the power output from any system device may impose a risk if the system
is not adequately “strong” or if it does not have adequate fast-acting voltage support to
maintain the bus voltage. Increasing power flow without adequate voltage support can lead
to local voltage collapse, and in severe situations, cascading voltage collapse.
This is a problem that is explicitly considered in the active power behavior and recovery of
wind plants and HVDC systems. Normally in weak grids, the active power recovery is
deliberately kept slower than the reactive power recovery (per the preceding paragraph) for
the reason that the power often cannot be evacuated from the wind plant or HVDC terminal
to the loads until the voltage has been restored. Thus, in weak systems, the active power
recovery tends to be slower. This is fundamentally at odds with the challenge facing South
Australia, where fast power recovery is necessary for FFR.

An assessment of the power recovery performance of FFR devices under consideration would
need to be undertaken by EMTP-type simulation analysis of the South Australia grid. This
“weak system” performance analysis could be performed as part of the post-fault recovery
analysis discussed above.

4.4! Differences between Under-Frequency and Over-Frequency Response

The ability of a generator to provide an over-frequency response is different from under-
frequency response. Over- and under-frequency response entail different costs, can have
different speeds, and can provide different levels of aggressiveness. As such, these two
responses should be disaggregated and considered two different products or services, similar
to up-regulation and down-regulation in some ISO’s.
Figure 43 shows the droop characteristic for a plant. Consider a plant operating at the
intersection of the dashed blue (nominal frequency) and dashed green (dispatched set point)
lines. As frequency increases above the deadband, the slope of the droop characteristic tells
the plant how to decrease output. It is noted that if the plant is operating at a minimum
generation level, then the plant may not be able to decrease output unless it is decommitted.
As frequency decreases below the deadband, the slope of the under-frequency droop
characteristic tells the plant how to increase output. Note that the plant needs to be operating
below its maximum output in order to have headroom to increase output.
Over-frequency response has no opportunity cost and little or no variable cost. It is beneficial
to the system in the event of a loss of load event, during some system separation events, or in
response to overly aggressive UFLS on a loss-of-generation event. This last consideration is
not insignificant, as the discussion of backswing of frequency illustrates. To the extent that the
cost to enable fast over-frequency response is very small, enabling this response at all times
should be considered.

109!
! !
!
The aggressiveness of the response (droop) may vary for over-frequencies versus under-
frequencies. In the past, this response was symmetric. This was largely a historical artifact of
mechanical (e.g. flyball) governors that were symmetrical (the weight was fixed). But modern
controls, and particularly those available on wind or PV plants, need not be symmetrical.
Droop characteristics may be different for over versus under-frequency for wind, and
especially PV (since there are no mechanical stress limits on how fast PV can back down). The
system should take advantage of these responses.
While the primary focus of PFR tends to be on under-frequency events, as was noted above,
historically over-frequencies can lead to, or be part of, severe system disturbances (e.g.
Florida’s Turkey Point blackout in the USA, Malaysia blackout). Unlike combustion-based
resources (especially gas turbines), wind and, especially solar PV, can reduce power output
fast and to very low levels without much risk of tripping offline. Such aggressive response
requires higher gains (less droop) for over-frequency. The recommended approach is to
explicitly allow (not require) that generating resources can have asymmetric droops and that
they can be less than 5%, based on mutual agreement of the grid operator and the plant
owner.

Plant Plant!Power!Maximum
Power =!rating!for!synchronous!plants
=!available!for!wind!plants)
Headroom

Plant!Power!Dispatch
=!actual!production

Droop

Deadband

System!short System!needs
of!generation power

100%!(Normal!Frequency)

Speed!(for!synchronous!machines)
Frequency!(for!wind!plants)

Figure 43 PFR droop characteristic.

4.4.1! Risk of Conflicts with PFR Droop Response

It is important to remember that this figure is static, and the response is linear. It is possible
to create or tune the dynamic response of generation such that the ability of the unit to
reasonably follow this droop characteristic is degraded or defeated. For example,
functionality that focuses on achieving specific power ramp rates or rapid response to market

110!
! !
!
signals can conflict with droop control. Ill tuned controls can result in frequency correction
overshoots, or in sustained frequency errors. For example, in the US, generator load controls
that defeat (“withdraw”) PFR are presently in widespread use. These cause poor frequency
response, especially in the US Eastern Interconnection. Efforts are currently being made to
correct this problem (by NERC). For any system to evaluate FFR properly, dynamic models of
PFR need to correctly capture all these effects.

4.5! Implications and Practicalities of Zero Synchronous Inertia Operation

While the focus of this document is limited to technologies that will limit RoCoF and hold
frequency, the trends of South Australia mirror those of other places on a trajectory towards
minimal and even zero synchronous generation. This is the subject of intense interest in
research circles, as evidenced, for example, by a session at the July 2016 IEEE Power & Energy
Society General Meeting dedicated to the topic. Broadly stated, the trends in development are
towards inverter-based resources that tend to mimic synchronous machines, in that they
synthesize an internal voltage phasor that sets both magnitude, angle and frequency. The
example shown in Section 3.3, Figure 24, illustrates that 100% inverter based operation of an
island is possible. In that case, a single device, specifically designed for this purpose, provided
all the needed functionality for the island. The on-going developments are centered around
how to get devices to share duty, especially current; how to get coordinated frequency setting;
and how to deal with greatly reduced fault currents and different fault current behaviors for
protection. Rapid development over the next few years is anticipated.
However, it is important to note that there are good reasons for the present design and control
of the voltage-sourced current controlled inverters that are predominant today for both wind
and solar PV generation. These devices make economically effective use of the relatively
expensive power electronic components; they tend to naturally coordinate well (between
many devices); they tend to have stable and relatively benign interaction with other system
components, including generally better damping of system oscillations. For the inverter-
based power generation industry to make wholesale movement towards equipment that
more closely mimics synchronous generation has non-trivial cost and systemic performance
issues that need to be addressed. This is not just a simple question of adjusting control
algorithms.

4.5.1! WECC and other Research on 100% Inverter-based Systems

There is good research being done in this arena. Of particular note, a new paper117 looking at
a hypothetical future US Western Interconnection with 100% inverter based generation shows
great promise for this (virtual synchronous machine) approach. Some good work is coming
out of National Grid in the UK.

117!Transient!Stability!Analysis!of!an!all!Converter!Interfaced!Generation!WECC!SystemU!Deepak!Ramasubramanian,!Vijay!Vittal,!
John!M.!Undrill.!IEEE!GPM!2016!

111!
! !
!
A new initiative in the European Union, called Massive InteGRation of power Electronic devices
(MITIGATE), includes participation by 13 countries, 12 TSOs and is funded for about 17M Euros.
The scale of the project (and funding) gives a sense of the breadth of the challenge.

4.6! European Rooftop Solar Experience

Figure 44 Overspeed event due to loss of load of 5 GW in Europe resulting in

disconnection of 10 GW of DER at 50.2 Hz119
Germany has extensive experience with retrofitting DER to comply with updated connection
requirements. About 12,700 MW of rooftop solar was connected to Germany’s low-voltage
distribution system. During 2005-2006, it was not anticipated that there would be high
penetrations of DER, so the connection requirement was established such that the rooftop
solar inverters must trip offline for over-frequency events at 50.2 Hz120. The European grid
was designed to survive a sudden loss of load of only 3,000 MW, much smaller than the
amount of generation that could be lost upon common-mode tripping of rooftop solar. Unlike
voltage, frequency is, for the most part, the same across the interconnected power system. A
loss of load (such as loss of an intertie during export conditions) could increase frequency to
50.2 Hz, which could result in much more than 3,000 MW of rooftop solar (depending on solar
resource at that time) common-mode tripping. The dashed red trace in Figure 44 shows a
simulation of such an event in the European grid, with the rooftop solar systems tripping offline

119!Kastle,!G.!and!Vrana,!T.,!!http://www.cired.net/publications/cired2011/part1/papers/CIRED2011_1275_final.pdf!!
120!J.!Boemer,!et!al,!“Overview!of!German!Grid!Issues!and!Retrofit!of!Photovoltaic!Power!Plants!in!Germany!for!the!Prevention!of!
st
Frequency!Stability!Problems!in!Abnormal!System!Conditions!of!the!ENTSOXE!Region!Continental!Europe,”!1 0International0
Workshop0on0Integration0of0Solar0Power0into0Power0Systems,!Aarhus,!Denmark,!Oct.!24,!2011.!

112!
! !
!
at 50.2 Hz and then trying to reconnect after 30 seconds. The green trace simulates the same
event, but with the PV providing over-frequency droop response.

Nine hundred distribution system operators had to come together to reach agreement to
retrofit some 350,000 existing systems at an estimated $520 M cost to correct this reliability
problem121. Being proactive and undertaking careful planning is one of the lessons learned
from the German experience.

There are two issues at play here. One is that the existing connection requirements should be
revised so that DER supports system security – so that DER does not common-mode trip in
response to a small voltage or frequency deviation. The second is whether DER should be
required to provide FFR capability. The new Australian standard AS4777, which came into
effect this year, requires frequency response from distributed PV. Presumably, the continuous
nature of this new standard will avoid the problems created by the discrete control
requirement in Germany. There is hysteresis in AS4777, which should avoid cycling risks.
Complex dynamic events, involving both high and low swings of frequency may create
unexpected behaviors. But whether DER should be capable of under-frequency response is a
more open question. On the one hand, DER is growing quickly, and as Germany has seen,
retrofits can be expensive, time-consuming, and resource-consuming. It is unlikely that
rooftop PV owners will be interested in pre-curtailing their systems to try to make money in a
potential FFR (or PFR) market. However, if one believes that very high DER penetrations or
micro-grids are realistic future scenarios, then proactive planning would suggest that
requiring FFR in the connection rules for future small resources might be prudent. But great
care must be taken to avoid unintended consequences.

121!B.!Ernst,!SMA,!“Evolution!of!LV!PV!Interconnection!Requirements!in!Germany,”!Utility!Variable!Generation!Integration!Group!
Spring!Technical!Workshop,!Anchorage,!AK,!May!2014.!

113!
! !
!
5! INTRODUCTION: FFR REQUIREMENTS AND ANCILLARY
SERVICE SUITABILITY
The situation in South Australia, while extreme, is not unique. The decline in synchronous
inertia and displacement of traditional sources of frequency sensitive generation that is
accompanying the addition of large amounts of wind and solar generation to bulk power
systems is the subject of considerable concern in the industry. Increasingly, entire grids are
faced with the prospect of operating under conditions where the majority of system
generation is from inverter-based, i.e. wind and solar, generation. Additional resources, such
as energy storage and HVDC, are growing. As the traditional sources of primary frequency
response (PFR), i.e. governors on the turbines of synchronous generators become scarcer
under some operating conditions, new sources of frequency response are emerging. Much
has been written about inertia-based controls for wind turbines, fast acting load controls,
battery energy storage and a diverse host of other technologies and concepts.
For system planners and market designers, evaluation of the system frequency for these
conditions is new ground. As described in 2.1, it is well understood that the initial rate of
change of frequency (RoCoF) increases (becomes more negative) with dropping synchronous
inertia. Indeed, the calculation of initial RoCoF is one of the few quantitative relationships that
is relatively simple to provide in closed form. But beyond this, the relationship between
traditional PFR resources and new resources is complex. In addition to PFR, the resources,
described in depth above can provide FFR, which can be more effective than a similar quantity
of PFR. Again, in general, it is well understood that with declining synchronous inertia the time
available for frequency responsive resources to deliver the arresting power necessary to
create an acceptable frequency nadir declines. However, the diversity of behavior from the
emerging resources presents a quantitative challenge. The planner and market designer
must “have enough” of these resources available to meet the system needs.
Several systems, notably ERCOT and EirGrid, have made inroads into understanding the
problem of “how much is enough?” Engineers in those, and other, organizations have
spawned the introduction of a variety of fast frequency response ancillary services. In the US,
ERCOT has lead development of understanding of the relationships between system inertia,
PFR and FFR. In this section, we use a very simplified model of the South Australia system to
illustrate the analysis necessary to answer questions like “How much is needed?” and “What
is most effective?”

5.1! Arresting Power and Energy

For the majority of power systems, including the major interconnections of North America,
frequency response requirements in the first seconds following major system disturbances
drive primarily towards one overriding objective: avoiding involuntary UFLS.
To that end, great attention is being given to understanding and assuring that sufficient
arresting power is provided to stop frequency decline before specific levels of under-
frequency are broached. The fundamental concepts and needs have been well documented.

114!
! !
!
The NERC Frequency Response Initiative Report122 provides arguably the definitive (at least
for North America) discussion. A number of studies by the authors123, 124, 125, as well as other
work in the industry have shown that a variety of frequency responsive technology beyond
traditional reliance on governor response from synchronous generation can be highly
effective in providing this essential arresting power.

5.1.1! Frequency Nadir: the metric of primary frequency response

There are a variety of complexities, but overall a single measure of performance captures how
well a system is performing: the depth of the frequency nadir following a disturbance. For
most systems there are specific credible (“design-basis”) disturbances for which involuntary
UFLS must be avoided. Various other measures are used to help determine how well the
system is meeting that objective. For example, settling frequency and frequency response
are, respectively, a proxy for how well the frequency was stabilized and an indicator of how
various parts of the system contributed to the outcome. One measure that gets lots of
attention is RoCoF. RoCoF, for most systems and major interconnections is, in itself not a
performance objective, but rather a measure that is useful to help understand what behavior
(especially speed) is needed.
The focus here is exclusively on the ability of resources to improve frequency nadir following
loss of generator or infeed disturbances. Most of the concepts presented should extend to
RoCoF objectives, in the rare system where managing RoCoF is itself a legitimate performance
objective. (Discussion of the vulnerability of power system components to high levels of RoCoF
is addressed in a separate report126).
For discussion of FFR, with particular attention to South Australia, Figure 5 from above is
reproduced here in Figure 45, with additional annotations on the FFR. In the figure, the total
power signature of a resource providing FFR is broken into four distinct regions. The energy
delivered before reaching UFLS is noted in green. The only arresting energy that is useful
towards avoiding involuntary UFLS must (obviously) be delivered before the system frequency
reaches that threshold. Arresting energy delivered after that time, shown in orange, is still
useful in managing the depth and timing of the frequency nadir. In systems with staged UFLS,
the distinction between the green and orange becomes less distinct. Ultimately, the modelling
presented here is based upon the objective to avoid involuntary UFLS, and the results
presented here are based on the concept of keeping the frequency nadir above a known
threshold. The third area, in blue, represents the period during which the FFR should be being
replaced by PFR. This must be done slowly enough that the frequency doesn’t drop again.

122!NERC!(2012c).!Frequency!Response!Initiative!Report.!Draft.!September!30.!http://www.nerc.com/docs/pc/FRI%20Report%209X
30X12%20Clean.pdf!!
123!N.!Miller,!et!al,!“California!ISO!(CAISO)!Frequency!Response!Study,”!Nov.!9,!2011.!https://www.caiso.com/Documents/ReportX
FrequencyResponseStudy.pdf!!!
124!N.!Miller,!et!al,!“Eastern!Frequency!Response!Study,”!NREL//SRX5500X58077,!May!2013.!
ttp://www.nrel.gov/docs/fy13osti/58077.pdf!!!!
125!N.!Miller,!et!al,!“Western!Wind!and!Solar!Integration!Study!Phase!3!–!Frequency!Response!and!Transient!Stability”,!NREL/SRX
5D00X62906,!December,!2014.!http://www.nrel.gov/docs/fy15osti/62906.pdf!!
126!New!RoCoF!advisory!report!by!GE.!!Planned!release!December!2016.!

115!
! !
!
And finally, for some resources, most notably wind inertia-based FFR, the arresting energy
provided must be recovered by reducing the total power delivered to a level below pre-
disturbance.

Figure 45 Time Elements of FFR

116!
! !
!
6! ILLUSTRATION OF SOUTH AUSTRALIA FFR REQUIREMENTS
AND ANCILLARY SERVICE SUITABILITY

In this section, we provide an analytical demonstration of the efficacy and requirements for a
FFR service in the context of meeting AEMO system performance requirements. The intent of
this work is to quantitatively show the relationships between synchronous inertia,
conventional primary frequency response services and possible new fast frequency response.
The analysis uses a very simple and approximate system representation, and is intended to
give results that are quantitatively meaningful, but not precise. The intent here is illuminate
processes by which in-depth analysis and more detailed models can be used to plan the
system and to guide procurement of ancillary services and operation as it relates to frequency
response.

In no way, should the reader regard this illustrative analysis as a substitute for detailed, high
fidelity modeling and analysis of the system. Nor should the reader regard this as an exercise
in replicating any specific event in South Australia, or validation (or invalidation) of the present
UFLS practice. We are attempting to look forward here.

6.1! Discussion of Model and Events

This section describes very simple representation used for the investigation. The model was
developed based on inputs from AEMO, and assumptions made by the authors.

6.1.1! South Australia Loadflow Model

A sketch of the system representation is provided in Figure 46. Equivalents of thermal and
wind generation, and South Australia system load are indicated. Since primary frequency
response is such a critical element to the issue, the thermal plant equivalent is segregated into
an equivalent of the units providing primary frequency response (i.e. those which have
governors enabled and which will respond to a drop in unit speed), and those that are not
responsive (i.e. either have governors inactive or which are at maximum power or valves wide
open). The distinction is important, and must be correctly made in actual system studies. The
transmission system is largely ignored, except for an explicit equivalent of the Heywood
Interconnector, as shown. The MurrayLink power injection is separately represented, with a
highly simplistic power injection (i.e. there is no dynamic model of the DC included). As noted,
the model is intended to give insight into frequency response.

117!
! !
!
 Figure 46 Sketch of Illustrative System Model

6.1.2! South Australia Dynamic Model

The dynamic model of the system is a fundamental frequency positive sequence
representation, of the type commonly used for system planning on software platforms like
PSS/e or PowerFactory. The cases shown here were run on the PSLF software platform.
Representation of the synchronous generation includes standard 6 state machine models and
excitation systems. As noted, a governor model is included on the responsive thermal
equivalent. The wind turbine equivalent uses a single representation of a typical, present
commercial utility scale wind turbine generator (WTG), with no frequency sensitive controls
enabled (unless otherwise noted below). The load model includes voltage sensitivity and a
dynamic component based on standard load modeling used in the US Western
Interconnection planning studies. The load is represented in granular pieces, to allow for
approximation to the UFLS system in place in South Australia.

6.1.3! November 1, 2015 Event

To calibrate the simple model, it was tuned to give a plausible response to the November 1,
2015 separation event. The loadflow was adjusted to approximate the grid condition before
the event, as shown in Figure 47. The system inertia in South Australia is modeled at 9000
MW-sec for the event. The AC import for the case, and therefore the size of the event for
simulation of the event, is 232 MW on the Heywood Interconnector.

118!
! !
!
 Figure 47 Loadflow for November 1, 2015 event cases
The turbine governor and underfrequency load shedding was tuned to give a reasonable
approximation to the measured event. A comparison of the simulated response and a
measurement of the event is shown in Figure 48.

119!
! !
!
 Figure 48 Measurement and Simulation of November 1, 2015 Event
The match is quite good, and provides a basis on which to do the illustrative work. The reader
is cautioned to remember that tuning a model to a single case does not constitute validation
of the model, which is considerably more difficult.

6.2! Requirements for South Australia: Present Low Inertia System

This section examines requirements for frequency response in the South Australian system as
it is operated most of the time today.

6.2.1! Recent History for South Australia

AEMO has shown that there has been a general trend of dropping synchronous inertia in the
South Australia system over recent years. The trend has continued in 2016. Figure 49 shows
data provided by AEMO up through September 26 of this year, presented as a duration curve.
The curve shows inertia down to a bit below 3000 MW-sec.
In this section, performance and behavior down to this level will be examined. Later, the
investigation will look at even lower levels.

120!
! !
!
 2820 MW.s

Figure 49 Recent (2016) History of Inertia in South Australia

The severity of frequency events is dominated by the amplitude of the generation-load
imbalance. For this discussion, that corresponds to the amount of power being imported to
South Australia on the Heywood interconnector. A duration curve of the AC imports (this does
not include the flow on the DC MurrayLink) is shown in Figure 50. During the period reported
here, the flow on the interconnector exceeds 600MW about 2.3% of the time, and exceeds the
level at the time of the November 1, 2015 event about 56% of the time.

Figure 50 Recent (2016) Imports on Heywood Interconnector

121!
! !
!
6.2.1.1! Correlation between Inertia and Imports
Since the simultaneous condition of high import and low inertia represents the more
challenging case, it is of interest to check on the correlation between high imports and low
inertia. Figure 51 shows a scatter plot and correlation trend line for the data of the preceding
two figures.

Figure 51 Inertia and Imports Correlation (2016)

The correlation is almost nil, with a slightly positive sign. The result indicates that no
reasonable inference of system inertia can be made from the import flow levels. If mitigation
is designed that is keyed to the inertia levels, other means to estimate system inertia will be
required.

6.2.2! Context using example of November 1, 2015 event

For the purpose of understanding the requirements and benefits of providing FFR, it is useful
to set aside the present autonomous underfrequency load-shedding scheme, and examine
how the system would behave without it.

In Figure 52, the model is run for exactly the same initial conditions and event as the calibrated
reference case, but with all the load-shedding – both voluntary and involuntary – blocked from
acting. As would be expected, the system frequency swings much further, with the frequency
nadir being lower and later.

122!
! !
!
 Figure 52 Simulation of November 1 Event without UFLS
Since a considerable amount of attention is given to inertia, it is instructive to consider, in
isolation, the effect of altering inertia. Figure 53 presents a thought exercise, in which all other
parameters are kept fixed, but the inertia is adjusted by modifying the MVA of synchronized
unresponsive generation. Again, this is for the November 1, 2015 size event; i.e. tripping of the
interconnector when loaded to 232MW.
The results are directionally as one would expect: for lower inertia, the initial RoCoF is higher,
the frequency nadir occurs sooner and is deeper. And, conversely, adding inertia improves
and delays the nadir. It is of some interest to note that even at greatly increased inertia, this
particular event still results in UFLS. The orange curve is for inertia roughly twice the highest
level presently observed in South Australia (per Figure 49 – maximum inertia with the
exception of a few hours of outliers is around 12,000 MW-s).
The learning from this is useful, but limited: Committing generation for the sole purpose of
adding inertia to improve the frequency nadir and avoid UFLS is not sufficient. If added
commitment of synchronous generation adds to the available primary frequency response,
then the results look different, as will be shown below.

123!
! !
!
 Figure 53 Frequency Response for Varying Inertia

These relationships are illuminating in two ways. They show:

1.! The effect of changing inertia because economics drives different levels of inertia from
synchronous generation in South Australia
2.! The efficacy of deliberately changing (or limiting) inertia, outside of the economic
dispatch and commitment. In other words, how effective is adding inertia compared
to other means?
This shows that adding inertia alone is relatively ineffective means of avoiding UFLS. If the
objective of adding inertia is to buy time for PFR and increase the likelihood that UFLS will
successfully act to disconnect load to avoid dropping below 47Hz, then the reduction in RoCoF
accompanying increased inertia will be beneficial.

The results of this sequence of cases can be mapped to show the relationships between
system inertia and other performance metrics. Figure 54 shows how the nadir gets deeper
with decreasing inertia (blue trace). If the amount of PFR is increased, all other things being
equal, the nadir improves, as shown in the purple trace. In that sequence, the amount of PFR
is doubled, by making twice as much South Australian thermal generation responsive. The
time available to invoke other actions, besides UFLS, is shown with the green points: for this
size event and operating condition, the threshold of UFLS at 49Hz is crossed in one second,
whereas more than 2 seconds passes with inertia at 9000 MW-sec. The upper reaches of this
curve are beyond the levels of inertia that South Australia sees today, but are illuminating in a
historical context, as South Australia in recent years past would operate at these higher levels.

124!
! !
!
 Figure 54 Inertia and Primary Frequency Response Relationships
This sequence of cases sheds some light on how the system behaves without FFR or UFLS.
The primary focus of this work is to consider what can be done with FFR. For this discussion,
we focus on avoiding any UFLS.
Figure 55 shows time traces of example cases in which discrete FFR action is taken at 0.5
seconds after the tie line trip event starts. The amount of PFR is held constant in this set of
comparisons – in the sense that the units providing PFR are the same for each case and are
at the same dispatch. FFR is generic here, in the form of power injected into the South
Australian grid and sustained past the frequency nadir. The amount of FFR (i.e. the power
level) is selected to make the frequency nadir 49Hz; that is the least amount necessary to avoid
UFLS. As expected, the amount required decreases as inertia increases. In the upper trace,
with inertia of 3000 MW-sec, the change in RoCoF is clearly visible when the FFR “hits” at about
49.4 Hz.

125!
! !
!
 Blue: reference case with UFLS

Red: FFR cases without UFLS

Time (seconds)

Figure 55 Illustration of FFR Requirement and Inertia

The results of this sequence of cases is presented in Figure 56. The red curve (which uses the
right-hand axis) shows the power needed versus system inertia, and the blue curve shows the
total energy injected between actuation and the frequency nadir. At first look, the fact that
the required energy increases may be counter-intuitive, but the time until the frequency nadir
is longer for the heavier system, and more energy is needed to arrest the decline of the heavier
grid. Said differently, the lighter system requires a faster sharper application of arresting
power, but it is easier to “catch” the lighter system.
126!
! !
!
 These are all for a 232MW event

Figure 56 FFR Requirement to Avoid UFLS

This curve allows for comparisons of the efficacy of the FFR. First, we can consider the impact
of FFR power rating on nadir, i.e. how much does the next incremental MW of “generic” FFR
impact the frequency nadir? The blue trace in Figure 57 shows the marginal efficacy of the
FFR. As discussed above, the FFR is more effective at arresting the decline of a lighter system.
For example 1 MW of FFR produces 13mHz of improvement in the nadir (i.e. better margin to
UFLS) at 3000MW-sec, compared to 9mHz of benefit at 9000MW-sec.

These are all based on a 232MW event

Figure 57 FFR Power Sensitivity

For a calibration exercise, in the figures it can be seen that, at inertia of 3000 MW-s (H = 3000),
130 MW of FFR is needed to avoid UFLS (for this 232 MW event). At H = 6000 MW-s, 75 MW is
needed. This gives a basis for comparison: What does it cost to increase the inertia by
3000MW-s, compared to the cost of adding 55 MW of FFR? The answer to the 1st question is
complex, and requires a production simulation to be answered with any fidelity. But, some
calibration can be made. Consider 5 alternatives:
127!
! !
!
 •! 75MW of controlled voluntary load-shedding: cost is a few ones of A$M probably, but
load shedding has societal costs.
•! 75MW of IBFFR from wind turbines: inexpensive (but some cautions apply; see below)
•! 75MW of BESS: say, A$75M, plus some variable operating costs.
•! 3000MW-sec of flywheel or condenser: say, A$100M, plus some variable operating
costs.
•! 3000 MW-sec of forced commitment of existing generation: say 500MW at
A$30/MWhr out-of-merit/uplift cost = A$15K/hour. For (say) 3000 hours/year =
A$45M/annum. [Obviously this type of calculation MUST be done with a security
constrained production simulation. Considerations for production cost simulations are
discussed below]
We will revisit this exercise below, with consideration of extremely low inertia and a more
generalized construct.

6.2.3! Sensitivity to Event Size (i.e. Initial Heywood Loading)

The results presented so far are based on a single large event: loss of 232MW of infeed. The
question arises “How linear is this information?”

The next sequence of tests are against the event size; e.g. the loading on Heywood. In Figure
58, higher loading on Heywood is tested. The pair of cases are for the same inertia, no UFLS
and the same PFR. The event is 50% bigger (~348MW), and RoCoF is roughly 50% worse, but
the nadir is about twice as deep. Clearly, for this set of boundary conditions, the frequency
excursion doesn’t increase linearly with event size.

Figure 58 50% increase in event size

Closer inspection of the behavior of the system shows that the behavior is relatively linear, if
viewed from the right perspective. In Figure 59, loci of frequency nadir versus the amount of
FFR provided are shown. In these cases, the FFR is again a generic block switched at 0.5
seconds. The amount of FFR required to achieve a specific frequency nadir is very linear: the
128!
! !
!
slope of the three traces is essentially the same. In Figure 60, for the same conditions, the
minimum FFR required to avoid UFLS (i.e. keep the nadir up to 49Hz), is plotted. The amount
of required FFR is very linear with the amount of loss-of-infeed. But extrapolation of the curve
(dotted red) does NOT go through zero. This particular PFR “covers” approximately the first
~165MW of loss.
This test is for a fixed PFR and fixed H (9000 MW-sec) (and still no UFLS). The result suggests
that for a given inertia (H) and PFR, the amount of required FFR is equal to infeed minus an
amount of PFR “coverage”. It is also of some interest to note that the time to frequency nadir
shortens as the event gets bigger, but at these (now) relatively high inertias, the difference in
timing is small. These results are quite linear, but simulations on a detailed, high fidelity model
might show a higher degree of non-linearity.

Figure 59 FFR vs. Frequency Nadir for different event size (fixed PFR and H)

129!
! !
!
 Figure 60 Minimum FFR necessary to Avoid UFLS (for one inertia and PFR condition)

6.2.4! Other Factors: Short Circuit Strength and Dynamic Performance of Synchronous
Generation
The results presented above were deliberately constructed with minimal changes to other
system characteristics such as inertia and PFR. But as the amount of synchronous generation
drops, the contribution to system short circuit strength will drop as well. At the lower end of
the range of inertia for this sequence of tests, the dynamic behavior of the voltage was
observed to be rather poor. The model was tuned up for the condition, adding more power
from wind and reducing the short circuit contribution from the synchronous machine
equivalents. The equivalent exciter, and the voltage regulator on the equivalent wind plant
were tuned as well. In Figure 61, a comparison of the same event (348MW loss) and inertia,
with different dynamic models is shown. The newer case (red trace) has a lower frequency
nadir (than the preceding case: green in this figure, red in Figure 58. This is mainly because
the voltage was kept healthier during the swing. This may seem counter-intuitive, but the
sensitivity of the load to voltage has been observed in our other work to occasionally be a
significant factor for frequency stability investigations. Healthier voltage means more load
active power, which exacerbates the generation-load imbalance. Nevertheless, healthier
voltages are to be desired, and are an important part of maintaining stability.

130!
! !
!
 Figure 61 Effect of Other Parameters
It is nevertheless an important point, that the types of parametric illustrations provided here
cannot be meaningfully made too precise. Modeling matters, as do details of dynamic
behavior above and beyond frequency response.
The test to determine the FFR requirement to meet the 49Hz target is shown in Figure 62. It
takes an extra 10 MW of FFR on top of the 135 MW required to avoid UFLS in the 348MW loss
reference model case. The suggests, not surprisingly, that all calculations need some margin.

Figure 62 FFR Requirement with Modified Dynamic Model

131!
! !
!
6.2.5! How fast is fast enough?
In the discussions of section 2.2, we gave considerable attention to the challenge of rapid
detection and decision making. But, clearly there is a need for speed when attempting to
arrest the frequency decline for conditions of high power-load unbalance and low inertia.
Figure 63 shows the results of a sequence of cases with low system inertia (at 3000 MW-sec),
subject to a 232 MW loss of infeed. The three cases are for different delay times between the
initiating separation event and step actuation of 145MW of FFR. The case (red trace) for only
50ms represents that fastest time that we believe could conceivably be possible with the best
available technology. On the other end, the green trace is for 500ms of delay, which
represents a speed that is clearly possible with a range of different technologies and subject
to a good level of confidence that the decision to trigger is proper. The blue trace is for ¼
second of delay, and represents an ambitious, but practically achievable speed of actuation.
As might be expected, the very fast response gives the best performance, resulting in the
highest frequency nadir. The ½ second response, with 145MW of FFR, is sufficient to meet
the 49Hz target.

Figure 63 Efficacy vs Speed of FFR for Low Inertia

Figure 63 shows a diminishing return on speed, even for a “light” system. A faster FFR resource
will require a lower power rating, primarily because it has more time to deliver arresting
energy. Figure 64 shows a case (in red) in which the extremely (perhaps impractically) fast FFR
power level is reduced to the minimum level required to avoid UFLS. In this case the required
FFR power drops to 130MW from 145MW. So, triggering super-fast (50ms), saves 10% FFR
power, over slow (500ms) triggering. As discussed in Section 2.2, there is higher risk of making
an unnecessary trigger decision with shortened decision time.

132!
! !
!
 Figure 64 Speed vs Power for Minimum FFR (to avoid UFLS)

6.3! Requirements for South Australia: Future extremely low inertia system
The current trends in South Australia are towards even lower levels of inertia. In this section,
the investigation is continued down to extremely low levels of inertia: 1000 MW-sec (which
have been reported since this work began). This investigation deliberately does not extend
below that level, even though it is fully conceivable that the present market and operational
constructs could drive the system below 1000 MW-sec. Nevertheless, as discussed in Section
4.5, operation of large integrated power systems without any synchronous generation is
presently the subject of a variety of research projects. Neither the available modeling tools
(i.e. standard stability programs) nor presently offered inverter-based generation is suitable
for this challenge today. Whether presently available technology, and indeed installed
infrastructure, can be adapted to zero synchronous operation is an outstanding question.
In the tests of these very low inertia conditions, the MVA of wind generation is increased to
allow most of the load not served by imports to be served by wind. The rating of the remaining
generation continues to drop, so the drop in SCR and voltage support significantly affects the
performance.

6.3.1! Stability Risks

A few simulations were performed that qualitatively highlight some risks as the system inertia
drops to extremely low levels. In Figure 65, the exercise started in Figure 64 is continued. The
case with inertia of 3000 MW-sec and 130 MW of FFR triggered at 50ms is shown in blue. As
noted above, it is just sufficient to avoid UFLS. When the inertia drops to 2000 MW-sec (green
trace), there is a slight decline in the frequency nadir. But when the inertia drops to 1000 MW-
sec, the system suffers from a relatively dramatic stability failure. The failure is not simply due
to an inadequate amount of FFR, as the case in Figure 66, with FFR increased to 200MW also
fails.

133!
! !
!
On closer inspection, the very low H case fails were due, in significant part to poor voltage
control and angular stability. The case was statically unstable, that is, even without a
perturbation, the system could not operate as an island.
While the details of this rather simple simulation are not very quantitatively meaningful, this
failure is almost certainly reflective of risk that will be observed in the real system: The dynamic
behavior of the system with very low inertia and very low short circuit strength will be
substantially different. Diligent checks of stability are absolutely required. A requirement for
control improvements is likely.

Figure 65 Stability Problem at Extremely Low Inertia - First Example

Figure 66 Stability Problem at Extremely Low Inertia - Second Example

134!
! !
!
6.3.2! Dynamics of extremely low inertia system
On closer inspection, the extremely low inertia case failures were due primarily to poor voltage
control and angular stability. With improved tuning (but not extra equipment), the illustrative
case can be made stable, for loss of the Heywood interconnector. Figure 67 shows a few
successful trips of the Heywood interconnector for an inertia of 1000 MW-sec. In one case
(the blue trace) the interconnector was unloaded. Trip of the line is therefore not a cause of
power-load unbalance, but it is essential to understand that the operating condition after the
line opens is hugely different than before. The effective gain on every regulator (both
frequency and voltage control) is very different. There is no assurance that the system is stable
in this condition, as indeed was the case for the system tested in the previous section. Opening
the line causes a small, but not trivial perturbation in the system. In the other two cases, the
loading of the tie line was 180 MW, and FFR equal to the tie line loading is applied. In the green
trace, the FFR is the super-fast 50ms triggering, whereas the red trace is for triggering at ½
second. It is of interest to note that triggering an amount of FFR equal to pre-disturbance
loading on interconnector results in a nearly instantaneous arrest of the frequency decline.
Triggering ½ way to the UFLS threshold, still allows for successful recovery. The latitude for
mismatch of FFR with event size declines with inertia.

Figure 67 FFR equal to event size

6.3.3! Inertial-based FFR from Wind Generation

Before examining even larger events with extremely low inertia, we begin to look at the
possible contribution of FFR with more sophisticated controls. That is, with closed-loop
frequency sensitive controls, rather than the open-loop “triggered” behavior built into the

135!
! !
!
cases presented to date. To that end, the cases presented here are based on the inertia-based
FFR described at length in Section 3.2.

6.3.3.1! Tests of Wind IBFFR

An extension of the case with the lowest Inertia (H = 1000 MW-sec) is shown in Figure 68,. In
this new case, the blue trace, the wind generation has an IBFRR feature enabled. In this case,
there is 1000 MVA of wind generation producing about 720 MW of power. IBFFR is generally
limited to about 10% of the wind turbine production, so was assumed that about 80MW of
response will be available. Since this event is for 180MW of lost infeed, the case applies
100MW of switched FFR at 0.25 seconds.
The case has an interesting outcome: the frequency decline with IBFFR and 100 MW of FFR
has a successful arrest of the initial frequency drop, with an initial frequency nadir of about
49.4Hz; i.e. about the same as for 180 MW of FFR. But, the withdrawal (and backlash) of the
IBFFR is incompatible with PFR from the thermal plant. The primary frequency response does
not “fill in” fast enough to allow the IBFFR to enter its recovery phase. The system frequency
collapses. The case is a fail. For success, either the IBFFR needs to be sustained longer, or
PFR from the thermal plant needs to be faster. Making IBFFR faster is not obviously very
beneficial, which is consistent with earlier results. But, withdrawal could be delayed (within
limits). We investigate further below.
This first test, “off the shelf” control settings were used with the IBFFR (labeled IBFFR#1 in the
legend). In the case presented, control (red trace “IBFFR#2”) is made considerably more
aggressive and more sustained. The resultant frequency recovery is substantially better, but
the case ultimately fails again during the recovery phase. PFR from the thermal plant still
needs to be faster, and the result suggests that some tuning of PFR could coordinate with
IBFFR better.
The point of these cases is to show that there are options and risks with closed-loop control.
A further point is that IBFFR, and indeed many types of FFR, can have their behavior adjusted
or tuned for the best performance. This observation applies for both closed-loop and open-
loop controls. Broad questions of “does ‘it’ work?” are adequate to get guidance and
calibration, but as with all the system dynamics under these high stress conditions, details of
control matter greatly. This applies to design and application of any FFR, including that from
wind plants.

136!
! !
!
 Figure 68 Comparison of 2 IBFFR Controls
6.3.3.2! Heywood at 450MW and Extremely Low Inertia
The cases presented in the previous section are for relatively modest levels of import. In those
cases, the size of the event (180MW) was somewhat similar to the maximum contribution of
the wind IBFRR – i.e. about 10% of production.
Here we examine an event that is large relative to available IBFFR from wind. The result of
tripping the interconnector when loaded at 450MW are shown in Figure 69. There are three
cases shown. In two of the cases, only switched FFR is applied. When exactly the amount of
FFR (450MW) is triggered at 250ms (the red trace), the system survives. With a little less FFR
(400MW), the frequency swing drops below 49Hz (green trace). Addition of IBFFR (the blue
trace “WI”) greatly helps the initial event, but poor coordination with PFR causes a fail. There
are probably many ways to fix this, if there is adequate participation by the remaining PFR
resources. Nevertheless, this highlights the narrowing window to have exactly the right
amount of switched FFR triggered in a very short period. It is difficult, if not impossible, to
imagine that autonomous, granular UFLS could possibly be relied upon to trip just the right
amount of load. The addition of IBFFR clearly buys some benefit, and may allow other controls
to take over. But, at least for this model, sustainability is a problem: the PFR needs to fill in
faster.

137!
! !
!
 Figure 69 Illustration of Sensitivity to Balance at Low Inertia
In Figure 70, IBFFR is applied, combined with switched FFR equal to the lost infeed. In this case
(the red trace), the IBFFR from wind helps improve and stabilize response with switched FFR.
It does not “replace” the PFR (load balancing), but rather contributes significantly to the
stability of the system during the initial swings. Open-loop IBFFR would not be as beneficial to
stability. The contribution of the IBFFR could make otherwise insufficiently fast PFR “fast
enough”, although that is not the case in this specific example.

138!
! !
!
 Figure 70 IBFFR benefits - rebalancing dynamics
6.3.3.3! Heywood at Limit and Extremely Low Inertia
A final case, in which we drive the system to the maximum stress under consideration in this
report, is shown in Figure 71. In this case, the interconnector is loaded to 650 MW and system
inertia is at 1000 MW-sec. The switched FFR alone (blue trace), triggered at ¼ second was
very slightly late to avoid crossing the 49Hz threshold, but more important, the discrete action
did not coordinate well with the PFR. The addition of IBFFR (red trace) helped stabilize this
case, which by slightly speeding up the switched FFR to 200ms, avoids crossing the 49Hz
threshold. The case suggests that, at best, relying solely on existing continuous controls and
block FFR will not be robust. Note that this case, when run with a non-trivial fault before line
trip was extremely unstable (not shown). Maintaining stability for this extremely stressed
condition will be a non-trivial challenge. And of course, the extremely low SCR (out of scope
here), will need to be considered.

139!
! !
!
 Figure 71 IBFFR at Limit of Low Inertia and High Heywood Loading

6.3.4! Summary: Trade-off of FFR requirement with Inertia

Results of this extremely low inertia investigation have been added to the results with
moderate inertia (3000 MW-s and greater), and are shown in Figure 72. This figure ties
together much of the essential findings of the lengthy discussions presented above.

140!
! !
!
 Figure 72 Relationship of Inertia to Min FFR and Event Size

In the figure, the efficacy of adding inertia relative to that of adding FFR is given by the
difference between the red and blue curves. The time before the system reaches 49Hz without
FFR is noted with each data point. Under these conditions, for example at 450MW import (as
indicated by the blue arrow), it takes about 120MW more FFR to avoid UFLS when the system
inertia is 1000 MW-sec than it does when the system inertia is 9000 MW-sec. Under these
conditions, the FFR must act within about 250ms for the 1000 MW-sec condition, whereas this
result for the 9000 MW-sec is based on 500ms action. The need for very rapid response
disappears with higher inertia; this is the reason a 500ms action was used. The red trace, for
very low inertia, is a very simple relationship: the FFR needs to equal the contingency size. This
is because the PFR has so little time to act, that it essentially doesn’t enter the relevant
dynamics. The blue curve, with higher inertia, is displaced to the right. The slower RoCoF that
goes with this higher inertia, buys time for the PFR to act. Effectively, the higher inertia creates
the time necessary to get the equivalent of about 175MW of FFR from PFR. The slope of the
curve is not the same as the red curve. This is because larger events, for the same inertia,
result in faster frequency decline. The PFR is less effective, and therefore more FFR is required.
From a planning perspective, this allows some calibration: The capital cost to meet the
marginal requirement for FFR is the difference in cost of (about) 330 MW of 0.5 second
response FFR compared to the cost of 450 MW of 0.25 sec response FFR. A very rough
estimate for 120MW dedicated energy storage device(s) would be A$120M. This figure is
towards the high end of a range, since at least some of the necessary FFR could come from

141!
! !
!
relatively less expensive resources, like IBFFR from wind generation. As noted above, the cost
to add inertia (in the form of synchronous flywheels) is on the order of A$35/kw-sec, so 8000
MW-sec, would be roughly A$280M. Notice that this exercise assumes a degree of linearity
that may be at odds with the exact details of the system performance. As has been noted
(repeatedly), this simple model is insufficient to get precise results, nevertheless, this result
(combined with the results shown in Figure 53) suggests that system solutions that focus solely
on addition of inertia (in the form of flywheels or other additions that are solely evaluated on
their inertia benefits) are unlikely to be economically optimal.

6.3.5! Determining Economic Benefit of FFR

To determine the relative marginal economic efficacy of adding FFR compared to imposing
operating constraints in the form of constraints on South Australia inertia and imports from
Victoria, detailed comparative production simulations are required. Unfortunately, imposing
dynamic constraints on production simulations can be challenging. The results of dynamic
simulations need to be mapped to static boundary conditions that can be modeled in a
production simulation. There is industry precedent for this practice. For example, (in North
America) transient stability derived path ratings are often mapped into (so-called)
nomograms. Figure 72 a set of nomograms show how a production simulation could impose
the interrelated constraint of import limit on the Heywood interconnector with committed
synchronous inertia. The system is assumed to be constrained by “available response”. We
introduce this concept here, because from a variable cost perspective (i.e. the constraints of a
production simulation) willingness to shed load by UFLS (or other means) and the amount of
FFR available are equivalent and additive. The different traces are measures of the sum of the
allowed load shedding plus the amount of FFR provided. For clarity, consider the blue “no
response” trace. If we suppose that no load shedding is allowed and that no resources are
available to provide FFR, then this line represents the loading constraint on AC interface. At
minimum inertia (of 1000 MW-s), zero power can be imported. As inertia increases (e.g. by
commitment of synchronous generation in South Australia), then import limit eases, up to
about 165MW at 9000MW-s. Now consider the red “200 MW” line. If 200 MW of FFR is
available, then at minimum inertia, the AC import limit is 200MW. At 9000MW-s, import limit
is 330MW. But, 200 MW of load shedding would produce the same result, as would (say)
100MW of FFR and 100MW of load shedding. This is a production costing constraint. Other
measures, including those that might require capital expenditures (e.g. reactive
compensation, or improved controls), could be needed, but they have no bearing on the
variable cost of operation. Using pairs of simulations with this type of constraint allows a
meaningful calculation or comparison of what FFR (or UFLS) is “worth”, from a variable cost of
operations (and emissions) perspective. This approach would allow for the evaluation of added
inertia as well; i.e. if dedicated flywheels were added to increase inertia all the time, the
constraint would shift to the left, rather than up. A similar comparison of annual VOC would
show the energy and emissions value.

142!
! !
!
 Figure 73 Production Cost Constraints for Import and South Australia Inertia

This exercise is, of necessity, isolating the effect of inertia and FFR. It is clear from the results
presented above, that faster (more, better) primary frequency response would affect the
amount of FFR needed for the 9000 MW-sec inertia case, whereas, faster conventional PFR
response becomes less useful in avoiding UFLS for the 1000 MW-sec inertia case. Faster PFR,
as for example that shown below (in Figure 74) from wind turbines will serve to offset load
shedding. Also, there is a presumption that UFLS that acts at or below 49Hz will “save” the
system. This means that the frequency swing should not go below 47Hz, and the system
should not go black. For the low range of inertia, this almost certainly optimistic. That means
that the region of the figure in the upper left hand side needs to be mostly, if not all, FFR rather
than a mix of FFR and UFLS.

6.4! Wind with IBFFR and PFR control

The general context of this report is focused on options and requirements to provide FFR.
However, as the results shown throughout this section reemphasize, the need and efficacy of
FFR cannot be isolated from the amount and quality of primary frequency response available.
In Figure 74, a case in which the wind generation is dispatched (i.e. curtailed) to 90% of the
power available from the wind, and primary frequency response is enabled. For clarity of
understanding, it is important note that for this illustration we have assumed that wind speed
is higher and have not actually reduced the output of the wind plant. The control settings for
143!
! !
!
this case have less droop (higher frequency gain) than the typical 5% used (in North America)
for synchronous generation. So, this is for the wind providing primary frequency response
equal to 10% of the available wind. In this case, this corresponds to about 80MW of PFR
available from the wind plants. The combination of the inertia-based FFR plus the primary
response, allows for an equal (80 MW) reduction in the amount of block FFR (i.e. load tripping)
necessary to stabilize the system. This illustrates two important points: (1) primary frequency
response from wind plants is highly effective, and in this case, is substantially more effective
than the PFR from the thermal plant, and (2) PFR from wind works well in combination with
IBFFR. Again, as discussed in section 3.2.4.1, unlike the provision of IBFFR, provision of PFR
from the wind plant has an opportunity cost for the generator equal to the value of the energy
curtailed to enable the underfrequency response.

Figure 74 Illustration of Wind providing both IBFFR and Primary Frequency Response

6.4.1! Context to Other Systems

The size of the event under consideration in South Australia, i.e. up to 650 MW loss of infeed
for complete loss of the Heywood Interconnector, is large relative to the system inertia in
South Australia. Even at an inertia level of 9000 MW-s, the loss of 650 MW represents a ratio
of about 15:1, inertia to event. At 1000 MW-s, the ratio is less than 2:1.

144!
! !
!
For some context, it is interesting to look at some other systems. For example, the design
basis event for the US Western Interconnection (for which involuntary UFLS in not allowed) is
about 3000MW. The minimum inertia is greater than 300,000 MW-s, so the ratio of inertia to
event size is greater than 100:1.
In the isolated Texas system, the design basis event is 2750MW. In recent history, the lower
end of the inertia range has a typically been around 120,000 MW-s, but ERCOT anticipates this
to drop in the near future to around 70,000. In this case, the ratio of inertia:event size will be
around 20:1.
The calculation is not as simple in Ireland, as there isn’t a single event size. Rather they look
at the worst loss of infeed that can happen for each event. Nevertheless, the 500MW DC
infeeds tend to be the biggest limiting event. The system minimum inertia drops in the
extreme to around 20,000MW-s. So, the minimum inertia:event ratio is on the order of 40:1.

The point of this is further reinforcement that this challenge is outside of the present industry
experience for large systems.

6.5! Strategy for Extremes of Low Inertia and High Import

This review of technology and characteristics of the South Australia system suggests that a
well-designed SPS could meet the necessary requirements of system security and economy.
The severity of the problem to be addressed in South Australia combined with the substantial
risk of overswing/backswing problems associated with discrete control actions almost
certainly dictates that any discrete triggered FFR SPS continuously adapt to the exact level of
power/load unbalance that will occur on separation of South Australia from the rest of the
NEM.
The requirement for some sort of “master controller” that implements this function would
periodically monitor the potential loss of infeed, i.e. the flows on the AC tie or across interfaces
where the system can separate. The system would calculate the magnitude of the power/load
imbalance in South Australia from separation events. This would be an indication of the
magnitude of FFR and PFR response needed to stop the frequency decline and restore balance
between generation and load. The master controller would also need to track the amount of
FFR available from the various resources that could be triggered to act. The controller would
then determine which FFR resources should be triggered, and possibly to what power level, if
the ac intertie trips. This data would be updated every few seconds or minutes. Then, when
the tie trips, the controller knows exactly what FFR devices are required to respond and the
triggering signal can be sent out instantaneously. This is essentially a fast-acting control
scheme where the triggered action is continuously updated in advance to adapt to changing
grid operating conditions.
There is precedent for this type of situationally adjusted arming. For example, a similar
adaptation is used to arm the remedial action scheme (RAS, aka SPS) on the Pacific HVDC
Intertie in the western US. Also, the Intermountain HVDC system in the US was equipped with
a similar “contingency arming scheme” scheme that could adapt control actions to ten
different monitored system events.
145!
! !
!
The impetus for this type of “adaptive” triggered control action is because, as the illustrative
analysis above showed, the magnitude of the control action needs to be closely matched to
the magnitude of the disturbance. If the control action is too small, frequency would decline
too far and load shedding could be triggered. On the other hand, if the control action is too
large, the imbalance of generation and load would change sign and frequency would quickly
rise and overshoot 50Hz. An underfrequency event would inadvertently cause an
overfrequency event. By adapting the magnitude of triggered FFR to system operating
conditions, the frequency decline can be quickly arrested in a manner that will allow PFR
resources to handle the remaining unbalance between generation and load. The inclusion of
at least some fast acting closed-loop frequency sensitive FFR (including IBFFR) will increase
the robustness of the system, and aid maintaining stability.!

6.5.1! Special Protection Scheme summary

Specifically, in summary, we recommend consideration of an SPS that:
•! Is primarily based on fast load tripping,
•! Detects islanding quickly,
•! Is armed a priori to rapidly trip load “equal” to AC imports
•! Is augmented by other fast technologies
Some key summary points are:

•! For all but the very most extreme conditions, 250ms for action is fast enough to avoid
involuntary UFLS.
•! The consensus of (GE) experts is that a secure system with this speed is possible. Faster
might be possible.
•! The requirement for exactness of “equal” increases as RoCoF increases, which occurs
as:
o! import increases
o! inertia in South Australia decreases
•! Other fast technologies will help stabilize and improve robustness of such a scheme.
6.5.1.1! An SPS won’t be easy
There are significant challenges and functions of an SPS. To determine how much FFR, i.e.
load tripping plus other FFR must be armed at any instant of operation is complicated by:
What is left after separation?

•! Where does the system break?

•! It isn’t just Heywood at the South Australia-Victoria border; there are many places that
the South Australia system can break away from the rest of the NEM.
Distributed PV?

•! Are you tripping generation, too?

•! How much do you get by tripping?

146!
! !
!
6.5.1.2! The SPS may be needed for other risks
It is likely that the SPS can provide benefits for other classes of events. Most notably:

•! Overload scenarios of the general type that occurred on Sept 28, 2016 may be
addressed, at least in part, by such a scheme.
•! The required amount of load tripping (and other FFR actuation) will not be the same as
for complete loss of in-feed, but will likely have some similar dependencies.
•! Such an SPS might also be useful for managing partial loss of the interconnection as
well. Detailed design of the SPS should consider such other functions.

6.5.2! Role of other FFR technologies

The SPS must distinguish between Sustained and Transient FFR technologies:

•! Sustained FFR provides power until secondary rebalancing (e.g. unit start-up, resync,
manual load adjustments, etc.). It is a substitute for PFR.
•! Transient FFR “buys time” for slower PFR to act. (AEMO’s present 6 second product).
•! Provision of PFR from resources other than synchronous generation should be
considered. An example of wind generation providing PFR shows good performance.
Technologies discussed in this report that provide a sustained response (other than load
shedding) include:

•! BESS, curtailed solar & wind, possibly some other energy storage technologies and
possibly the HVDC.
•! Each MW of these technologies displaces 1 or more MW of load shed.
Technologies discussed in this report that provide a transient response include:

•! Includes inertia-based FFR from Wind turbines, high power/lower energy storage (high
inertia synchronous condensers, ultra-caps). A short duration HVDC FFR function is
more reasonably regarded as a transient resource.

6.5.3! Requirements on other FFR Technologies

Speed of response

•! Devices should be able to respond quickly.

•! Analysis suggests that extremely fast response (e.g. less than 100ms) is not required,
and may not produce incremental improvement over responses on the order ¼ second
(even slower for the present system with 3000 MW-sec inertia).
•! Closed loop response; with minimal deadband and over-frequency response, will be
highly desirable. Control should be stable and well behaved for backswing over-
frequencies.

147!
! !
!
Transient FFR

•! Careful consideration to the timing of withdrawal and recovery (if any; e.g. for inertia-
based FFR for wind turbines) is needed.
•! For IBFFR, response should begin to be withdrawn at the frequency nadir; but not too
quickly (assuming that response can be sustained that long).
•! Management of component duties within the HVDC will be needed in the event that
FFR is designed to temporarily overdrive the system. These are likely to be the limiting
factor in FFR from this resource.

6.6! Autonomous Schemes and Backswing Risk

Autonomous, distributed UFLS is difficult to get exactly right in most systems. Finely grained
and staggered UFLS are the standard approach to attempting to have the right amount of
load trip for different events. Such a system, with many small steps, separated by a
combination of different delays and frequency thresholds, requires time to self-coordinate.
With the potentially very wide range of inertia and RoCoF, it will be even more difficult to get
right in South Australia…at least using local measurements. A system that arms autonomous,
distributed UFLS based on current (real-time) risk (i.e. loading of Heywood), would help. This is
a similar concept to that outlined above, but with the distinction that arming would be set by
a centralized system, but triggering would be based on local measurements. This approach
is, in some regards simpler and probably less expensive. However, since there is a multiplicity
of separation scenarios, such arming is necessarily approximate, since local measurements,
unlike a complete SPS, will be unable to differentiate different events.
This inevitable imperfection raises the risk of tripping too much load, causing a backswing of
frequency to over 50Hz. Around the world, backswing has caused many system-wide
blackouts. In Figure 75, an illustration of the risk is shown. Note that this just a drawing, and
not a simulation. The call-outs point out the sequence. New AS4777 may help mitigate risk of
backswing by reducing power generation by distributed resources, but it also has risk of
exacerbating loss of generation. Note that this is not necessarily a problem, but rather that
consideration of the contribution of AS4777 compliant resources should be included in
simulations intended to evaluate the risk. The big unknown and risk in this scenario is
generation tripping during the high frequency. Causes could include lean blown out (LBO) of
gas turbines, action of protection, or sensitivity to high positive RoCoF. The point of this
illustration is that too much open-loop response, including excessive UFLS, is a significant risk.
And conversely, fast continuously acting resources will add robustness and reduce the
requirement for precision in tripping just the right amount of load.

148!
! !
!
 Figure 75 Backswing Risk Illustration

149!
! !
!
7! FFR AND SIR ANCILLARY SERVICE SPECIFICATION
In this section, we present an example of the mechanism by which AEMO can determine the
necessary amount of the new FFR ancillary service to procure. The intent of the exercise is to
show the relationship between synchronous inertia, the existing (or modified) primary
frequency response (PFR) ancillary service and a new FFR ancillary service. As shown in the
preceding section, the physics of the problem dictates that these three aspects are inexorably
intertwined, and the most effective and economic procurement of ancillary services will need
to consider all three.
The present nascent industry practice for FFR products has created a somewhat limited
definition of FFR that tends to limit or exclude technologies that could be effective in meeting
AEMO’s objectives (and in opposition to one of the stated objectives of the statement of work:
“…specification should facilitate the broadest possible participation…”)
To facilitate procurement from the widest range of technologies, we illustrate a process by
which FFR resources with a range of response characteristics can be normalized to a
reference FFR functionality, (e.g. a block of arresting energy). This provides a mechanism by
which different characteristic resources can bid and be compensated for equivalent beneficial
impact on system frequency performance. This concept is an extension of the basic
underlying SIR-FFR-PFR trichotomy.

The idea is to create relationships that can be used in the co-optimized market clearing of
energy and ancillary services. The underlying assumption is that introduction of too many
ancillary services will be hard to police and to achieve liquidity in the market. Thus, the
expectation is that multiple characteristic resources will be able to bid into a single (or few, if
necessary) products, and be “normalized” in their systemic value to AEMO.

7.1! Power, Energy and Timing

Most of the systems investigating FFR, have tended to have rather strict definitions of the
expected behavior of FFR. This has the potential to limit the technology choices, since strict
adherence to proscribed power signatures may exclude some technologies. Further, it is by
no means assured that the prescribed performances provide the most benefit to the grid.
In this context, the questions of “what behaviors produce the best outcomes?” and “how does
one compare resources with very different behavior characteristics in their abilities to
maintain system security?” become pressing. The concepts and results presented in this
section present a broader framework for characterizing the quantitative impact on system
security, particularly frequency nadir, for any resource that contributes arresting power. The
intent is to allow “apples-to-apples” comparison between resources, so that compensation,
via markets or other means, can confidently:
• Reward the behavior that best meets the reliability need,
• Allow the broadest participation by resources that can contribute,
• Ensure that the right amount of resources is procured, and
150!
! !
!
• Pay fairly for the benefits delivered.

7.2! Evaluation of FFR Performance

In the following sequence, we present a construct for evaluation of FFR. To that end, the
performance of the South Australia illustrative system without FFR is established as a baseline.
The baseline case is for a relatively high level of system inertia (12,000 MW-s), with the intent
of illustrating an approach that will be useful across all of the NEM and not limited to just South
Australia.
A sequence of tests, in which a 250MW, 250ms “block” of idealized FFR energy is injected at
successive ½ second intervals (the total energy injected is therefore 62.5 MJ). The concept is
illustrated in Figure 76, with the baseline shown as the blue trace, and one perturbation. The
perturbation is the red trace, in which the energy pulse starts ½ second (500ms) after the
event starts and ends after another ¼ (250ms).

Figure 76 "Ideal" FFR perturbation test

The added FFR arresting energy improves the frequency nadir, and delays the time at which
the nadir occurs. This occurs, to differing degrees, for each interval up to the time of the nadir.
Obviously, FFR injections that occur after the nadir have no benefit (arguably by definition). A

151!
! !
!
composite of all the cases, run in a sequence up to four seconds into the event is shown in
Figure 77.

Figure 77 Frequency Response for Trip of Heywood Interconnector

7.2.1! FFR Response Efficacy Mapping

The sequence of cases results in differing impact on the frequency nadir, even though the
arresting power and energy are exactly the same for each perturbation. The change in nadir
is shown in Figure 78.

152!
! !
!
 Figure 78 Nadir Improvement vs. Timing of Arresting Power Injection
This result is then linearly mapped, or normalized to reflect how any arresting energy applied
at the given time will impact the frequency nadir, as shown in Figure 79. (The green is
extrapolated data; the efficacy tends to roll off as the nadir approaches, partly because there
is less scope for improvement as the frequency approaches the nadir. The extrapolation is
unimportant for the point of this illustration. When the transfer function is actually calculated
with a complete system model, the perturbation simulations should be run to times past the
nadir.)

Figure 79 FFR Efficacy Mapping

It is significant to note that injection of arresting energy later in the event has much higher
efficacy at improving the nadir compared to injection that occurs immediately after the event
starts. This is due to the dynamic interaction of the FFR with the PFR. By allowing a more
severe initial RoCoF, the PFR controls act more aggressively and produce better outcomes.

153!
! !
!
This sensitivity rather dramatically shows that faster response of FFR is not necessarily better.
In this case, injections after 1 second are more effective. The authors have tested this
approach on larger, more complex systems, and have found this trend. Broadly, we have seen
that in systems with frequency nadirs occurring several seconds after the event have the best
efficacy about one second before the frequency nadir. The behavior for a system with very
fast nadirs will have different quantitative results.
This may all seem rather convoluted, but the implications for most large systems are
significant: very fast response isn’t necessary, and taking time to make good decisions about
deployment of FFR is not only desirable from a robustness perspective, but uses FFR resources
with finite energy more effectively.

7.2.2! Evaluation of FFR with a Complex Response

In this section, a test case using inertia-based FFR on wind generation (synthetic inertia) is
used to illustrate the concept. In the case, the reference disturbance is applied. The sum of the
change in power (due to the inertia-based control) for the (equivalent) controlled wind plant is
shown in Figure 80.

Figure 80 FFR Power with inertia-based FFR from Wind Generation

The arresting energy for each interval of efficacy mapping for this FFR power signature is
bucketized and presented in Figure 81.

154!
! !
!
 Figure 81 Aggregate Wind IBFFR Arresting Energy

By adding up or convolving the per unit expected benefit (Figure 79) with the amount of energy
in each bucket (Figure 81), the impact of each “slice” of FFR energy can be determined, as
shown in Figure 82. By adding up the components, the cumulative expected impact on
frequency nadir is obtained, as shown Figure 83. So, for this example the expected
improvement in frequency nadir is 438 mHz.

Figure 82 Expected Component Benefit of FFR Energy

155!
! !
!
 Figure 83 Expected Total Benefit from IBFFR Energy

7.2.3! Confirmation
The frequency response of the time simulation that produced Figure 80 has an actual
frequency nadir improvement of 432mHz (0.0086pu), which is essentially exactly as predicted
(438mHz) by the linearized method. This provides an encouraging proof of concept, at least in
so far as showing that the impact of any FFR signature can be predicted, at least when the
expected impact on frequency nadir is relatively small.

Figure 84 Confirmation of Impact of IBFFR

156!
! !
!
7.2.4! Non-Linearities and Other Complexities

This method has been tested for much larger individual FFR resources, and as might be
expected, linear extrapolation of performance for large signals is imperfect. Tests on large
signal (i.e. lots of arresting energy) suggest that the method is rather conservative. That is, this
approach tends to somewhat understate the benefit of large FFR arresting energy injections.
From a system security perspective, this is highly desirable, as the risk of under-procuring FFR
services based on the linear approach appear to be small.

There appear to be promising approaches to improving the method, even for large-signal use.
The perturbation method provides useful insight into the impact of different FFR resources on
the timing of the nadir, as well as the amplitude. Experiments suggest that this information is
useful in refining the estimation of frequency nadir benefit. Further, use of this approach to
target other behaviors, such as RoCoF, looks promising. The approach is not a substitute for
complete dynamic simulations to validate performance.

7.2.5! Applicability: How High is the RoCoF?

The concept presented here is simple in a world where FFR is the ONLY resource active until
the nadir. No coordination with the speed of the PFR is necessary. In other words, FFR does
ALL the arresting, stabilizes the system until the PFR activates and takes over. This is the case
in systems that have extreme frequency declines (e.g. greater than 1 or 2 Hz/sec…i.e. like South
Australia).
But the more common situation is of systems with slower, more typical frequency declines.
These systems have historically used only PFR to provide arresting power, but may now find
themselves with marginally increasing RoCoFs and marginally short supplies of arresting
power. When considering FFR to augment existing (but dwindling) PFR, there is significant
value in considering the interaction between the two.

7.2.6! Conclusions of FFR Comparison Method

A method is proposed here, in which dissimilar sources of fast frequency response can be fairly
compared. A few key elements emerge from this work:
• Don’t forget that the primary objective is to improve the frequency nadir.
• Create an environment where every technology can contribute, and
• Be evaluated on its systemic benefit.
• Faster isn’t necessarily better, so don’t base requirements or market signals on that
assumption.

Next steps: This exercise is illuminating, but there are many complexities to be further
considered, including

• How linear are the results?

• How many different conditions need to be considered in order to get robust and
workable rules?
157!
! !
!
• How to keep rules up-to-date?
• Are there unintended consequences?
• How might other objectives be included/rewarded?
Detailed simulations, on a validated system model, are required to answer these concerns.

7.3! A Possible Market Construct

The method outlined above suggests that it is possible to make fair apples-to-apples
comparisons of different and complex FFR response. At least when the individual responses
being compared are of modest scale relative to system and event size. The question remains,
“what to do with this from a practical market or procurement perspective?”
One practical approach would be to use simulation of generic FFR to determine how much is
needed to meet performance objectives. This could be as simple as determining how much
generic FFR is needed to raise the frequency nadir for the specific operating condition (e.g. as
determined by economic security constrained dispatch) for the design event (e.g.
interconnector flow) to a level that meets the required frequency standards with some margin
(e.g. as set by the rules).
Participants in ancillary service markets could offer amounts and prices for their dissimilar
services. The method here would allow the dissimilar services to be compared for their impact
on the frequency nadir. The value, and therefore the clearing rank, of the offers would be
normalized to cost per unit of improved system performance (i.e. $/!Fnadir). The market
would clear when enough FFR is procured to meet the expected system needs. Of course, a
full market isn’t needed to realize this benefit. The approach is equally applicable to other
longer-term procurement approaches (e.g. tenders or bilateral contracts).
The net result of this approach is that resources with more desirable signatures will get a
premium based on their better impact on frequency. But other resources, which might have
less ideal signatures, but lower costs, can also participate: More participants equals better
economy. Prescriptive rules for specific signatures, that will limit the participation of a wider
range of resources, and generally result in less effective use of power and energy ratings of
FFR resources, can be avoided.
For the immediate case of South Australia, as noted above, the relationships will tend to
become simpler, with resources that provide response fast enough to be useful for the
extremes of low inertia having similar value. Unlike the larger interconnected NEM, where the
relationship illustrated above will likely hold, South Australia may take a simpler approach to
procurement. To wit, there is no need to unnecessarily complicate comparison of speed and
shape of response in South Australia because FFR needs to be very fast. In fact, the primary
distinction of importance for South Australia will be between discrete, switched resources (i.e.
switched blocks of loads) and continuously acting resources (e.g. BESS, IBFFR, etc.). We
showed above that at least some response ought to be continuously acting with closed-
looped controls. A further distinction may be sustainability of the output; i.e. IBFFR probably
needs to be handled separately.

158!
! !
!
7.4! Procurement Discussion
It is first important to note that physics should drive FFR procurement: maintaining the security
of the NEM, especially the South Australian system, is becoming increasingly challenging. This
is emphatically a power system dynamics problem that requires good utility engineering
practice. Market constructs must be founded on this, and must have sufficient flexibility to
adapt to what will inevitably be a rapidly changing situation – both in terms of available
technology and details of the grid dynamics.
Provision of FFR from most resources that might be built for the specific purpose of providing
that service involves significant investment in physical equipment. For dedicated resources
like batteries, flywheels, etc., the capital investment is the dominant cost. The variable cost of
providing FFR is usually relatively low. Other resources, which might be able to provide FFR in
addition to their main function (e.g. IBFFR from Wind, or FFR from temporarily over-driven solar
PV), may be able to offer the service after a relatively modest capital investment. Some
resources, like batteries, might also incur opportunity costs if they are performing other
market functions, such as energy arbitrage127. Design of a market for procurement of FFR
needs to consider these realities.

Procurement by an annual (or periodic) clearing market (i.e. winning resources are obligated
to provide FFR service when called upon, in return for a fixed revenue stream, plus energy net)
can be effective. Tendered procurement, in which a call is issued for competitive supply of FFR,
can also work. Variable (e.g. Day-Ahead clearing) markets for FFR may face challenges for
price formation and stability.
Procurement of FFR resources can follow various paths. The character of the service suggests
that the following points apply:

•! It will be challenging to get a day-ahead or real-time market to work. Variable (e.g. Day
Ahead clearing) markets for FFR will face challenges for price formation and stability.
•! Procurement by an annual (or periodic) clearing market could work. In such an
arrangement, winning resources are obligated to provide FFR service when called
upon, in return for a fixed revenue stream. Payments could also include an energy
delivered element.
•! Tendered procurement, in which a call is issued for competitive supply of FFR, can also
work. This approach might be desirable when the FFR resource is largely separate
from other (e.g. energy) markets, and the resource becomes a “network” asset, like
transmission or control investment. Such an arrangement would work for bilateral
contracts as well.
•! The methods outlined in this report for determining the amount of FFR needed to meet
performance and economic objectives will provide guidance for establishing how
much is needed, and how much can be spent on that procurement (e.g. auction caps,
etc.).

127!Energy!storage!may!participate!in!energy!markets,!buying!at!low!prices!and!selling!at!high!prices.!!Reserving!power!rating!for!
FFR!could!result!in!lost!revenues.!

159!
! !
!
 •! A part of the market design work should include debating the manner in which loads
are involved in FFR. Should participation by loads in triggered response schemes (such
as an SPS) be voluntary and paid? Or involuntary and unpaid (as in the existing UFLS
scheme)?
The requirements for FFR will continue to evolve in all parts of the NEM, including South
Australia. Processes will be necessary to keep the technical evaluation of the frequency
response requirements current. An annual review, at least, is probably required.

7.5! Plant Connection Mandates

Some of the technologies discussed in this report, most notably IBFFR from wind generation
and some types of load control, involve relatively inexpensive improvement to equipment.
Several system operators around the world have implemented requirements (i.e. grid codes)
that mandate that wind plants have IBFFR functions to interconnect. This was discussed in
Section 3.2.5.

160!
! !
!
8! RECOMMENDATIONS FOR FUTURE ANALYTICAL WORK
The section includes recommendation for future analytical work, primarily simulations, that
will aid in making good decisions regarding FFR services.

8.1! Recommendations for Detailed Dynamic Analysis

8.1.1! General
•! Illustrative simulation results used in this work are not a substitute for detailed, full
scale dynamic analysis.
•! Most analysis can be performed on fundamental frequency positive sequence tools
(e.g. PSS/e)
•! At very low inertia and low short circuit current levels, at least some analysis to assure
control stability (e.g. EMT studies) is warranted.
•! Broadly, we recommend that the analysis included in the illustration presented in
Section 6 be performed on the “real” system. A few specific analyses will be of
considerable value, including the analyses that produced Figure 72 and Figure 73.

8.1.2! Class of Analysis/Cases

Combined Fault and Trip Events.

•! Events which cause islanding of South Australia are likely to involve transmission faults.
•! Dynamics of all systems elements are complicated by faults.
•! No types of generation can inject active power into deeply depressed voltages;
attempts to do so tend to exacerbate transient stability and transient voltage collapse
problems.
•! Realistic fault and trip to island events and realistic loss of generation (e.g. in South
Australia) events need to be included in the mitigation design process.
•! Issues of maintaining control stability for such extreme conditions are much more than
just high RoCoF.
•! Stability considerations are likely to dominate functional and equipment specification
of any remediation technologies.
Dynamic/Control Stability.

•! As noted, issues of maintaining control stability for such extreme conditions are much
more than just high RoCoF.
•! Dynamic stability must be assured. Therefore:
o! Frequency domain analysis is required.
o! Eigenvalues must be stable before, and at projected quasi-static conditions
following disturbances.
o! Linearization of non-conventional technologies is required, and can be a non-
trivial challenge.

161!
! !
!
 o! Modal participation and eigenvalue migration may be surprising under such
substantially different conditions.
•! Simulations that will support the design of the recommended SPS are needed.
Specifically, design of the SPS will require that:
o! The necessary speed of load tripping be established,
o! All events that can lead to islanding be identified,
o! Better understanding and verification of the requirement for exactness of
“equal” as a function of inertia and import be developed.
•! A centrally armed autonomous SPS, using local triggering (section 6.6) will need
simulations to establish whether robust performance can be obtained under low
inertia conditions for the possible spectrum of separation events.

8.1.3! Recommendation for Modeling for Detailed Dynamic Analysis

Good model fidelity is essential. Creating and maintaining high fidelity dynamic models is (and
must be) an ongoing process. It requires significant effort. Validation should be made:

•! Against observed events; against validated/confirmed field settings (e.g. for UFLS
relays; generator protection).
•! For staged, field tests (e.g. like NERC model validation testing128).
•! Of load response (this is difficult; a program of monitoring should be considered).
•! Of dynamic performance of all resources providing FCAS, with attention to assessing
how those resources respond to frequency error.
Calibration analysis

•! Illustrative analysis included several parametric variations that could be valuable if

performed on the full South Australia system model.
•! If RoCoF measurement and event detection is used as a main method for triggering
FFR (not our recommendation), frequency from real events should be examined (at
multiple locations) to determine the size of the sampling window to manage the
tradeoff between accuracy in event detection and time to respond.

8.1.4! Recommendations for other technical analysis and simualtions

Short circuit screening

•! Evaluation of grid strength and minimum short circuit strength for future conditions is
needed.
•! Screening for short circuit levels below those acceptable to equipment manufacturers
or below current technology tolerances are needed. These might point to the need for
EMT analysis

128!http://www.nerc.com/pa/Stand/Project%20200709%20%20Generator%20Verification%20%20PRC0241/MODX027X
1_clean_2012Sept11.pdf!

162!
! !
!
ElectroMagnetic Transients analysis

•! Depending on screening results, as future very low short circuit strength scenarios
become defined, EMT analysis of the behavior of specific inverter-based resources,
(including, but not limited to new FFR technologies) will be required.
•! Analysis for risk of poor control performance due to weak grid may be required, but
after basic fundamental frequency and frequency domain work has established
credible alternatives.
•! Detailed modeling of MurrayLink in a suitable platform (e.g. PSCAD) to determine the
possibility of using the project as a source of FFR should be considered.

8.2! Recommendations for Detailed Economic Analysis

8.2.1! General: Production Simulation
Evaluation of the economic trade-offs between various alternatives done based on high
fidelity production simulations will give the most solid foundation for decision making.
For South Australia, production simulations that:

•! Use real (or realistic) hourly wind and solar production

•! Correctly constrain operation of the AC interconnector subject to the inertia and PFR
provided by synchronous generation in South Australia (per discussion in Section 6.3.5)
•! Constrain operation subject to the allowable level of involuntary load-shedding
•! Correctly capture the contribution of various alternatives (e.g. new supplies of FFR and
improved PFR) in partially or completely relieving constraints
This type of simulation will provide reference values for the impact on the annual variable cost
of energy, impact on carbon and other emissions. This type of analysis is essential to
determining the relative value of different strategies. Strategies could include addition of
capital equipment, more conservative operation, relaxed reliability or performance
expectations. It is not meaningful to only ask the capital cost of adding different widgets!
Simulations of these types normally assume that economic responses will be competitive, and
that markets has sufficient depth to avoid market power. Additional investigation of the
robustness of price formation for the operating conditions calculated by the recommended
production simulations may be warranted.

8.3! Zero Synchronous Generation Future

As noted in Section 4.5, this a future condition that needs to be considered. Technology and
solutions are far outside the narrower scope of this FFR investigation, but the state of the art
is changing very rapidly. South Australia should, for now, proceed with the understand that
there should be no expectation today that it can survive, islanded from the rest of the NEM,
without at least some synchronous equipment in operation. To make zero synchronous
generation operation possible in the near future, investment will be required, some of which
may be in technology that is not commercially available today.
163!
! !
!
9! REFERENCES
In addition to footnotes throughout the document, additional references are provided here.

9.1! RoCoF and Frequency Response References

DNV-GL, RoCoF Alternative Solutions Technology Assessment – Final Report Phase 1, Report
No. 16011111, Rev. 005, 17 August 2015.
EPRI/DOE Handbook 2003
DOE/EPRI 2013 Electricity Storage Handbook in Collaboration with NRECA, SAND2013-5131,
July 2013.

Application of rate of change of frequency constraints for high wind penetration scenarios in
a small power system, George Ivkovic, Power Engineering Conference (AUPEC), 2013
Australasian Universities, Year: 2013, Pages: 1 - 5, DOI: 10.1109/AUPEC.2013.6725374

Network Control System Protection Scheme: Design enhancement, Tien Ho, Power
Engineering Conference (AUPEC), 2013 Australasian Universities, Year: 2013, Pages: 1 - 6, DOI:
10.1109/AUPEC.2013.6725436

Advances in Wide Area Monitoring and Control to address Emerging Requirements related to
Inertia, Stability and Power Transfer in the GB Power System, D.H. Wilson et al, CIGRE Paper
C2-208, Paris, August 2016.

Smarter Network Storage: SDRC 9.7 Successful Demonstrations of Storage Value Streams, P.
Papadopoulos, A. Laguna-Estopier, and I. Cooper, UK Power Networks; March 2016; Pages 1-
71. http://innovation.ukpowernetworks.co.uk/innovation/en/Projects/tier-2-
projects/Smarter-Network-Storage-(SNS)/Project-
Documents/SDRC+9.7+Successful+Demonstrations+of+Storage+Value+Streams+LoRes+v1.p
df
J. Eto, et al. “Use of Frequency Response Metrics to Assess the Planning and Operating
Requirements for Reliable Integration of Variable Renewable Generation”, LBNL-4142E, Dec.
2010. http://www.ferc.gov/industries/electric/indus-
act/reliability/frequencyresponsemetrics-report.pdf

N. Miller, et al, “California ISO (CAISO) Frequency Response Study,” Nov. 9, 2011.
https://www.caiso.com/Documents/Report-FrequencyResponseStudy.pdf

N. Miller, et al, “Eastern Frequency Response Study,” NREL//SR-5500-58077, May 2013.

http://www.nrel.gov/docs/fy13osti/58077.pdf
N. Miller, et al, “Western Wind and Solar Integration Study Phase 3 – Frequency Response
and Transient Stability”, NREL/SR-5D00-62906, December, 2014.
http://www.nrel.gov/docs/fy15osti/62906.pdf
164!
! !
!
9.2! Frequency-Input Controls for Damping
Robust damping controls for large power systems, J. F. Hauer, IEEE Control Systems Magazine,
Year: 1989, Volume: 9, Issue: 1, Pages: 12 - 18, DOI: 10.1109/37.16744
Robustness issues in stability control of large electric power systems, J. F. Hauer, Decision and
Control, 1993., Proceedings of the 32nd IEEE Conference on, Year: 1993, Pages: 2329 - 2334
vol.3, DOI: 10.1109/CDC.1993.325613

9.3! HVDC
Implementations and experiences of wide-area HVDC damping control in China Southern
Power Grid, Chao Lu; Xiaochen Wu; Jingtao Wu; Peng Li; Yingduo Han; Licheng Li, 2012 IEEE
Power and Energy Society General Meeting, Year: 2012, Pages: 1 - 7, DOI:
10.1109/PESGM.2012.6345363

Description of the new RAS to be installed in the power system of Uruguay, D. S. Beledo; D.
Bonjour; F. Sanchez; N. Yedrzejewski, Innovative Smart Grid Technologies Latin America (ISGT
LATAM), 2015 IEEE PES, Year: 2015, Pages: 188 - 193, DOI: 10.1109/ISGT-LA.2015.7381151

Designing the hardware and software of the Pacific HVDC Intertie Remedial Action Scheme
using a programmable controller, M. R. Bhuiyan; A. E. Taylor; N. C. Kezman, Transmission and
Distribution Conference, 1991., Proceedings of the 1991 IEEE Power Engineering Society, Year:
1991, Pages: 790 - 796, DOI: 10.1109/TDC.1991.169594
Latest control and protection innovations applied to the Basslink HVDC interconnector, M.
Davies; A. Kolz; M. Kuhn; D. Monkhouse; J. Strauss, AC and DC Power Transmission, 2006. ACDC
2006. The 8th IEE International Conference on, Year: 2006, Pages: 30 - 35, DOI:
10.1049/cp:20060007
The Intermountain Power Project 1600 MW HVDC transmission system, C. T. Wu; P. R.
Shockley; L. Engstrom, IEEE Transactions on Power Delivery, Year: 1988, Volume: 3, Issue: 3,
Pages: 1249 - 1256, DOI: 10.1109/61.193910
Benefits of HVDC for reducing the risk of cascading outages and large blackouts in AC/DC
hybrid grid, M. Benasla; T. Allaoui; M. Brahami; A. Boudali, Control, Engineering & Information
Technology (CEIT), 2015 3rd International Conference on, Year: 2015, Pages: 1 - 6, DOI:
10.1109/CEIT.2015.7233170
Thyristor valve for the 12-pulse converter for the Champa-Kurukshetra HVDC transmission
scheme, Mohammad Hassan Jodeyri; Andrzej Dzus, 2013 IEEE Innovative Smart Grid
Technologies-Asia (ISGT Asia), Year: 2013, Pages: 1 - 6, DOI: 10.1109/ISGT-Asia.2013.6698790

MurrayLink, The Longest Underground HVDC Cable in the World, I. Mattsson et al, CIGRE
Session B4-103, 2004

165!
! !
!
Power Modulation of Sidney HVDC Scheme, part 1, RAS Control Concept, Realization, and Field
Tests, RK Johnson et al, IEEE Transactions on Power Delivery, Vol.4, No.4, October 1989.

166!
! !
!