http:d2cigre.

org

CONSEIL INTERNATIONAL DES GRANDS RESEAUX ELECTRIQUES
INTERNATIONAL COUNCIL ON LARGE ELECTRIC SYSTEMS

STUDY COMMITTEE D2
INFORMATION SYSTEMS AND TELECOMMUNICATION
2013 Colloquium
November 13-15, 2013
Mysore KARNATAKA - INDIA


* IIIT-Bangalore, 26/C, Electronics City, Hosur Road, Bangalore-560100 India
Fax: + 91 80 4140 7704 e-mail: sunil.vuppala@iiitb.ac.in

D2-02_07
REAL TIME ENERGY MANAGEMENT IN SMART GRID
by
Sunil Kumar Vuppala*, G.N. Srinivasa Prasanna
International Institute of Information Technology Bangalore (IIITB)
(IN)
SUMMARY

The issue of sustainability and eco-friendliness at smart grid causes a challenge to match
demand with supply at different levels. We need to improve the existing power grid
infrastructure by utilizing intelligent grid-state-aware optimal generation, distribution, and
loads. We model the smart grid energy management as a feedback system and analysis of its
dynamics from stability, efficiency and user comfort viewpoints. In this paper, we discuss the
modelling methodologies to handle large scale power grid problems, solution methods for such
large problems and interpretation of these solutions to estimate benefits. Modelling of energy
management as the price-setting demand-response mechanism is discussed.

Since even a 1000-customer problem each having 10 appliances with hourly varying time slots
and 10 supply side options results in a mixed integer linear program (MILP) with millions of
variables, the grid wide optimization is computationally intractable, and convexification
techniques and heuristics have to be used.

The model to be presented is an extension of standard MILP formulations, to incorporate
aggregation for computational tractability, and extensions to handle constraints such as user
comfort, emission limits, and appliance level constraints. The Comfort criterion limits the
maximum waiting time to service customer requests. The emission constraint limits the total
generation, and possibly its split into categories/locations. A novel feature of our work is
modelling of the appliance at a detailed level (major electrical characteristics; power flow,
response times, DR variables) and mutual inclusion and exclusion of appliances as constraints
and complex inter dependent appliance constraints.

A major feature of this model is the incorporation of uncertainty, in an intuitive fashion, using
substitutive, complementary and general constraints. The model used is an extension of robust
optimization, incorporating information of theoretic concepts to estimate the number of future
scenarios covered by the optimization.

http:d2cigre.org

CONSEIL INTERNATIONAL DES GRANDS RESEAUX ELECTRIQUES
INTERNATIONAL COUNCIL ON LARGE ELECTRIC SYSTEMS

STUDY COMMITTEE D2
INFORMATION SYSTEMS AND TELECOMMUNICATION
2013 Colloquium
November 13-15, 2013
Mysore KARNATAKA - INDIA


D2-02_07 2/9

Convexification is used to make the problems tractable, with targeted near real-time response.
The convexified model is solved by nested simulation and optimization using Cplex both at
consumer and utility levels. Run time of the solution is few seconds at consumer level and few
minutes at utility level.

A practical implementation is carried out by connecting the legacy appliance infrastructure with
controllable plug load sensors to participate in smart grid activities. We observed benefits of up
to 40% of energy saving during the implementation at a campus level network. The result
suggests that the model is to create mutual benefit among the consumer and the utility
companies with a flat demand curve during a day and the automation of the energy
management.

KEYWORDS
Smart energy management, smart plugs, demand response, energy conservation, smart meters,
real time analysis of power grid

1. INTRODUCTION:
The issue of sustainability and eco-friendliness at smart grid causes a challenge to match
demand with supply at different levels. Even though energy management problem is not new,
the concepts of smart grid make it more challenging. Smart grid is a framework for combining
the electrical power infrastructure with modern digital communication networks and
information technology. Two way communication, customer participation, renewable energies
and storage are important features of the smart grid [1]. The smart meter connects the consumer
at home or building and utility company as an interface.

The operation of utility and consumers is usually separated. However, we insist that there is
possible solution to financial and social advantages for both if they operate together. Energy
and demand side management are some of the programs for their interaction. The energy
management refers to the process of monitoring real time energy data, controlling the operation
of appliances, and conserving energy in a building or city. It is important to consider the
operation time and appliance type for energy management, because of the category specific
dynamic real pricing (RTP), introduced by utilities. The smart grid requires an important role of
consumer in order to achieve energy efficiency and demand response activities. The simulation
is conducted based on the assumption of two way communications which is one of the smart
grid features and installed smart meter for energy measurement. The response from consumer
after solving the optimization model transfers to the utility company as a feedback.


D2-02_07 /9
3
2. RELATED WORK AND CONTRIBUTIONS:
The development of wireless sensor networks, smart appliances and meters causes new
optimization problem as it allows monitoring and controlling to the fine grain level (till the
appliance level). The energy and demand side management programs are able to be beneficial
for both utilities and consumers by interactive and automation. A real time demand response
model proposed by authors in [2] minimizes energy bill of the consumer or maximize consumer
utility but they considered only one consumer case with simple utility function. Authors in [3]
targeted to minimize both energy bill and waiting times with optimal residential power
consumption scheduling framework but they have not considered inter appliance constraints.

Even though earlier work in [2, 3, 4] represents few aspects of energy management models, this
is the first time to the best of our knowledge to take into account for feedback of consumer in
real time, in order to set the price coordinating to DR. The previous models lack the scale of
problem and different consumer types.

2.1 Contributions in the paper:

The key contributions in our paper include:
Modeling of smart energy management system (as shown fig 1) including price
uncertainty and analyzes on the dynamics with the stable, efficient, and user-friendly
perspectives.
Detailed level modeling of appliances, containing complex inter dependent appliance
constraints.
Simulation, implementation and analysis of the model, and the results to measure the
mutual advantages of consumers and utility companies.

3. SYSTEM MODEL & ARCHITECTURE:
3.1 Optimization Model:
The system model includes the characteristics of electric appliances, consumers and utility.
Several features such as the comfort, demand, appliance constraints and participation to DR
program affects the bill of consumer. Our goal is to minimize this energy bill of consumer and
to maximize utility companys profit. Maximum waiting time to respond to a consumer is
restrained by comfort criterion. This is a new development that we include the appliances at a
detailed level and those mutual inclusion and exclusion as constraints. The price uncertainty is
considered in this model. So, the computation of energy bill/profit is divided into two parts
actual bill till the present time slot and estimated bill based on predicted price and demand for
future time slots.
The detailed mathematical model with objective, constraints are as follows:
Objective function: Minimize energy bill: (number of units * price of unit at time t) +
(predicted number of units * price predicted at time t)
(1)

Maximize Profit: Revenue (demand *price charged) - Operating cost

(2)



+
t
t
ac c
t
t
t
ac c
t
e e Minimize
, ,
~
*
~
) ( * t t

= = =
+ =
M
j t
t t
j
N
i t
t
ac c
t
i
N
i t
t
ac c
t
i
t t
c s d d OC R Maximize
1 1
,
~
1
,
)
~
( ) ( t t
D2-02_07 /9
4
TABLE I
GLIMPSE OF CONSTRAINT SETS IN MODELING
-Scheduling vector for timeslot t and appliance a
w - Waiting cost parameter ; C- segment of consumers (C1, C2, C3)
N = Total number of consumers (Integer)
M = Total number of generation options (Storage, renewable) (Integer)
P power quantity (kW) ; E total energy quantity (kWh)
price; d demand; e- energy ;
Max/Min Max/Min limits; lim limit;
ac- appliance category;


(3)
Comfort constraints:
cost of waiting with maximum waiting time;
A Set of appliances
t timeslot of the day; T-total time slots
Start time and end time of the job

(4)
Real time pricing with predication for category
c and appliance category 'ac with regulatory
conditions



(5)
Inter appliance constraints:
BM Large number
y binary constraints
n- number of product terms in SoP form
m-num of Boolean var in terms
max min
,
a
h
a a
a
h
h
a
x A a
E x
a
a

|
o
s s e
=

=
(6)
Demand constraints:
Demand to be met per appliance
Appliance power limits


(7)






Uncertainty min-max constraints with bounds
- Robustness parameter
applying strong duality
dual variables
t
ac c,
, o |



Equation (5) illustrates the general constraint specification method defined by consumer for
multiple appliances for validity or invalidity in our model. For example, when Appliance A
runs, Appliance B cannot work; Appliance types C and D always operate together etc which
can be represented in a truth table form. Equation (6) represents the demand constraints and
equation (7) represents uncertainty formulation using robust optimization. Convex polyhedral
uncertainty sets are used to specify many kinds of future uncertainties. These are linear
constraints that model microeconomic behaviour such as aggregate, substitutive and
complementary behaviour for energy demand of appliances.

Some examples of substitutive constraints are:
dem_a0 + dem_a1 + dem_a2 <= 50
dem_a0 + dem_a1 + dem_a2 >= 10

Complementary constraints are:
dem_a0 - dem_a2 <= 15
dem_a0 - dem_a2 >= 2
where dem_ax are uncertain demand variables.
t
ac c
price
,
t =
t
a
x
] , [ ; 0
...
lim
*
1
a
et
a
st
t
a
w
A a
a
et
a
w
a
st
a
w
a
a
w
t
a
x
T
t
t
a
w
>
e s s
s
=
} 1 , 0 {
1
* )] 1 ( | [
1
1
e
=
>

=
=
r
n
r
r
m
q
r q q
y
n y
y BM m z z
(

+
. .
t
ac c
t
ac c
t
ac c
t
ac c , , , ,
, t t t t
0 ; 0
. . .
) (
,
, ,
,
> >
> +
+ I

t
ac c
t t
ac c
t
ac c
t
ac c
e t s
Min
o |
t o |
o |

I s

^
,
, ,
| |
t
ac c
t
ac c
t
ac c
t
t t
I
D2-02_07 /9
5
This is more general formulation of Berstimas model [11]. In special cases without break points
and liner objective function, it allows to use strong duality and the min-max problem can
reduce to linear programming (LP) and results in improvement of solution time.

3.2 System Architecture:

Smart energy management system architecture is depicted in fig-1. Utility can serve 1000s of
buildings. Utility has a smart energy manager and each building can have a smart energy
managers. The smart meter at any building acts as an interface between the smart energy
managers at building and utility. The smart plugs, smart appliances in the building can
participate in the real time energy management.

Fig 1: Architecture of Smart Energy Management

The smart meters and smart plugs measure the real time energy consumption and this energy
usage information can be sent to utility for load management and billing activities. The real
time energy management at any building includes the optimization of energy bill subject to
specified constraints and consumer preferences and real time pricing. The optimal operation of
appliances, which allows the energy comfort trade off, is resulted by solving the optimization
model.
D2-02_07 /9
6
Each consumer operates a number of appliances. This model is tractable even by scaling the
numerous generators which serve number of homes/buildings suitable for area/city wise
distribution system. The optimization is a MILP, when the metric is linear. The MILP has a
great number of variables in the model based on the number of buildings served by the utility
company.

Utility can set the price using smart energy management system by considering the real time
information sent from the consumers and other parameters from energy markets. Based on this
information, utility can send demand response signals to individual consumers and smart
energy manager at buildings automatically calculates the load reduction possible based on
consumer preferences and specified constraints. This operates in nested simulation model.


4. IMPLEMENTATION & RESULTS:

We convexified the optimization model with objective function and constraints (Table 1).
Nested simulation is used to achieve the objective of the optimal operational schedule of energy
management. The efficient solution is possible by powerful solvers such as Cplex [5]. So, the
model is implemented initially using Cplex and later practically in an enterprise building using
smart plugs.

4.1 Cplex implementation:

Basic parameters considered for the implementation are from the reference [2]. We extended
further with 1000s of time slots and different cases are considered in running our model using
Cplex solver.

- Performance of the solution with uniform pricing:
This was run with 86400 variables, uncertainty of 10000 and 4000 timeslots. At timeslot=1,
it took 40000 iterations to solve the model. Then it was run for subsequent time slots.
Result of energy consumption is equal to maximum demand value (0.00083) in every time
slot. No bias is observed to any of the timeslot due to uniform pricing. So, the demand is
fulfilled within the requested timeslot itself. It may push some demand to neighboring
timeslot because of maximum hourly limits. Pricing does not have an impact on energy
consumption.

- Price is uniform for some time with in hour but vary across hours:
The model was run with 86400 variables with uncertainty 10000 and timeslot=1. Result of
energy consumption is varying over 24 hours between 0 to maximum demand (0.00083) in
every time slot. But within an hour it has not changed because of equal pricing within an
hour. No bias is observed to any of the timeslot within the hour.

- Price variations with continuous oscillation consecutively in the neighboring timeslots:
In this case, price goes up and immediately goes down and then goes up as a continuous
oscillation in the consecutive timeslots. Even energy consumption scheduling is also
oscillating within the 0 to maximum demand. This was run for 8640 timeslots with
uncertainty of 5000 and timeslot=1.



D2-02_07 /9
7

- Zero uncertainty in pricing case:
The robustness parameter ( ) can be varied. Result of energy consumption is equal to
maximum demand value (0.0083) in every time slot. No bias is observed to any of the
timeslot. So, the demand is fulfilled within the requested timeslot itself. Pricing does not
have an impact on energy consumption as it is a Zero uncertainty case. This was run for
8640 timeslots with uncertainty of 5000 and timeslot=1.

- 100% uncertainty in pricing case:
The robustness parameter ( ) is varied to achieve this case. Result of energy consumption is
equal to either 0 or maximum demand value (0.0083) in every time slot. It is observed that
the demand is fulfilled in the timeslots with lower price. This was run for 8640 timeslots
with uncertainty of 5000 and timeslot=1.

In last two scenarios, both the cases of oscillating prices and uniform price within hour
cases are tested and the same result was observed as earlier.

- Effect of vary ramp up, down limits:
The shifting of demand in consecutive hours is observed. More chance for optimization in
the system is observed if higher the limits of allowable ramp/up down. This was run for
8640 timeslots with uncertainty of 5000 and timeslot=1.

In summary from the experiments, we observed the stability of the system with uniform pricing
and continuous oscillation consecutively in the neighboring timeslots. In the first case, pricing
did not effect on energy consumption and in second case the energy consumption scheduling
was oscillating within the 0 to maximum demand. The sensitivity of the solution was identified
within reliable range and confirms it to be a stable solution when test performed with variation
of few parameters including price, demand ramp up and down limits. The analysis demonstrates
that our model is proper for real-time implementation. Response time was ranging from few
seconds to minutes, which are dependent on the operation of energy management at a building
or utility level.


4.2 Practical Implementation:

Sensors such as smart plugs were used at an initial practical implementation. The optimization
module gave the operational schedule and it was implemented (switch on/off) with the smart
plugs. Zigbee smart energy based [10] smart plugs was used for communicating the energy by
reading at regular intervals to a wireless gateway (sensing) and by receiving controls from the
gateway for automatic control of the attached appliances (actuation). The system was installed
in an enterprise building as shown in fig 2. The type of smart plugs used varies based on the
type of appliance (15amp / 30amp). In our case, gateway sent the data to the server and smart
energy manager is implemented in that server.

The gateway can communicate the sensing information to a server via local area network or to a
cloud server using GPRS/3G technologies. To implement similar solution in residential case,
the gateway can communicate to a server in the cloud. The smart energy manager can be
implemented in a server or in cloud based on the suitability and type of consumer.

I
I
D2-02_07 /9
8

Fig 2: Practical implementation with smart plugs

The pattern of energy consumption for various appliances (water heater, water cooler, printer
and coffee machine in an enterprise) is shown in fig 3. The pattern observed for several months.
The energy management for plug loads resulted in an energy conservation by up to 40%.
Automatically turning off the appliances and adjusting the load during peak hours suggested by
the optimization module saved the energy consumption. The system enabled to flatten the
demand curve over a day with the reduction of the peak load. Please note that we have
considered real time pricing with uncertainty in this module. Few more case studies are
described and analysed in our earlier work published in [6, 7, 8, 9].


Fig 3: Consumption patterns of break room appliances

5. CONCLUSION:

Modelling of the energy management as a price-setting demand-response mechanism is
discussed. Initial results indicate that our model is practical and scalable to an area wide smart
grid. This model is giving mutual benefit among the consumers and the utility companies with a
flat demand curve during a day and the automation of the energy management. In future, we
further develop this model to add renewable energies and continue large scale simulations.

D2-02_07 /9
9
BIBLIOGRAPHY
[1] The Smart Grid: An Introduction, U.S. Department of Energy, 2010. Online:
www.energy.gov
[2] Conejo, A.J.; Morales, J.M.; Baringo, L, Real Time Demand Response Model, IEEE
Trans on Smart Grid, Vol 1 no.3, pp: 236-242, Dec 2010
[3] M. Pedrasa, T. Spooner, and I. MacGill, Coordinated Scheduling of. Residential
Distributed Energy Resources to Optimize Smart Home Energy Services, IEEE Transactions
on Smart Grid, Vol 1, Sep 2010
[4] Mohsenian-Rad A.H, Leon-Garcia.A, Optimal Residential Load Control with Price
Prediction in Real-Time Electricity Pricing Environments, IEEE Transactions on Smart Grid,
Vol 1 no.2, pp: 120-133, Sep 2010
[5] IBM ILOG CPLEX [Online].
Available: http://www-01.ibm.com/software/in/integration/optimization/cplex/
[6] Sunil Kumar Vuppala, Kumar Padmanabh, Sumit Bose, Sanjoy Paul, "Incorporating
Fairness within Demand Response Programs in Smart Grid", 2nd IEEE international conference
on Innovative Smart Grid Technologies (ISGT 2011), Anaheim, California, USA, Jan 17-19,
2011
[7] Sunil K. Vuppala, GNS Prasanna, "Optimal Operational Schedule of Appliances with
Energy Comfort Trade off in Smart Grid", International Conference on Operations Research for
Sustainable Development in Globalized Environment, The 44th Annual Convention of
Operational Research Society of India, Kolkata, January 6-8, 2012
[8] Abhilasha Aswal, Sunil Kumar Vuppala, G N Srinivasa Prasanna,"Smart Grid:
Optimization Methods for Demand Response Systems", INFORMS 2011 Annual Meeting,
Charlotte, NC, USA, Nov 13-16, 2011.
[9] Sunil Kumar Vuppala, GNS Prasanna, Closed Loop Feedback System for Demand
Response in Smart Grid", International conference on Emerging Automation and Information
Technologies for power Industry Growth (PAC2013), CPRI, Bangalore, Apr 19-20, 2013
[10] Zigbee Smart Energy overview. Available online:
http://www.zigbee.org/Standards/ZigBeeSmartEnergy/Overview.aspx
[11] Bertsimas, D., Sim, M. The price of robustness, Operations Research 2004; 52(1);
pg:35-53