Large Wind Power Plants Modeling Techniques

for Power System Simulation Studies

C. Larose, R. Gagnon, G. Turmel, P. Giroux, J. Brochu, D. McNabb, D. Lefebvre

and number of wind turbines increase. Even if wind power

Abstract—This paper presents efficient modeling techniques plants are modeled as one equivalent wind turbine, the large
for the simulation of large wind power plants in the EMT number of wind power plants in the power system leads to
domain using a parallel supercomputer. Using these the same performance limitations. Furthermore, although
techniques, large wind power plants can be simulated in detail,
traditional transient stability programs are capable of
with each wind turbine individually represented, as well as the
collector and receiving network. The simulation speed of the dealing with large networks and large number of turbines,
resulting models is fast enough to perform both EMT and the natural electrical modes of the network are not
transient stability studies. represented. Consequently, studies such as interaction
The techniques are applied to develop an EMT detailed between series compensation and wind power plants cannot
model of a generic wind power plant consisting of 73 x 1.5-MW be addressed.
doubly-fed induction generator (DFIG) wind turbine. The ideal situation would be to perform fast EMT
Validation of the modeling techniques is presented using a
comparison with a Matlab/SimPowerSystems simulation.
simulations of the complete network in detail, including
To demonstrate the simulation capabilities using these wind power plants. The use of full EMT simulation models
modeling techniques, simulations involving a 120-bus receiving eliminates all uncertainties and simplification assumptions
network with two generic wind power plants (146 wind associated with traditional transient stability models.
turbines) are performed. The complete system is modeled using Development done in the past 25 years in real-time
the Hypersim simulator and Matlab/SimPowerSystems. The simulator is a key pointing in this direction.
simulations are performed on a 32-processor supercomputer
Recent supercomputers and real-time simulators now
using an EMTP-like solution with a time step of 18.4 µs. The
simulation performance is 10 times slower than in real-time, fully exploit the advantage of massive parallel processing.
which is a huge gain in performance compared to traditional Simulation of large networks can now be performed in real-
tools. The simulation is designed to run in real-time so it never time or close to real-time using EMT models. Moreover,
stops, resulting in a capability to perform thousand of tests via since the simulation is built to be running in real-time, it
automatic testing tools. never stops, and thousands of tests can be performed
quickly using automatic testing tools.
Index Terms—Modeling techniques, power system
This paper presents the modeling techniques used to
simulation, real-time, wind farms, wind power plants.
develop detailed models of wind power plants in the EMT
I. INTRODUCTION domain using a parallel supercomputer. The resulting
models allow simulation needs to be fulfilled in the time
F OR the next seven years, Hydro-Québec will be
completing the integration of 4000 MW of wind power
into its power system network. This new reality brings with
frame of EMT and transient-stability studies. Using these
techniques, large wind power plants can be simulated in
detail, with each wind turbine individually represented,
it a wide range of new simulation needs, especially when together with its collector and receiving networks.
dealing with an already complex network that includes The modeling techniques presented are: 1- average and
HVDC lines, asynchronous interconnections, SVCs and switching-functions modeling of power electronic
series-compensated lines. As a result, many power system converters; 2- integration of wind turbine models into the
studies, in the time frames of both EMT and transient wind power plant collector network, 3- the decoupling of
stability, must be conducted to properly integrate this new collector network equations using transmission lines with
form of power generation. the contribution of underground cable capacitances and
Unfortunately, the traditional tools used to conduct these wind turbine transformer leakage inductances.
power system studies are now reaching their performance These techniques are validated using a detailed
limits dealing with large wind power plants. For one thing, Matlab/SimPowerSystems (SPS) simulation. To
traditional EMT simulation offers precise results but the demonstrate the performance of the simulation using these
simulation time becomes extremely long as the network size modeling techniques, simulations involving a 120-bus
receiving network with two large wind power plants are
C. Larose, R. Gagnon , G. Turmel, P. Giroux and J. Brochu are with performed on a 32-processor supercomputer. Simulations
IREQ Hydro-Québec Research Institute, Varennes, QC Canada J3X 1S1 are performed 10 times slower than in real-time, which
(email: larose.christian@ireq.ca) (email: gagnon.richard@ireq.ca)
D. McNabb and D. Lefebvre are with Hydro-Québec TransEnergie, represents a huge gain in performance compared to
Montréal QC Canada H5B 1H7. traditional tools.
II. MODELING TECHNIQUES FOR LARGE WIND POWER PLANTS of the capacitor, this current can be integrated in order to
The development of accurate and fast-simulating models obtain the instantaneous capacitor voltage. The next figure
of large wind power plants for detailed EMT studies poses shows the implementation of the average model to represent
various technical challenges. First, all the turbines must be the back-to-back PWM converter of a DFIG wind turbine.
represented in order to study phenomena that can occur on
the collector network of the wind farm. Second, most
modern wind turbines comprise power electronic converters,
which require a small simulation time step and involve time-
Capacitor voltage
consuming matrix computations. Third, wind farm collector Idc1 calculations Idc2
Idc1=Pac1/Vdc Idc2=Pac2/Vdc
networks use very short underground (U/G) cables and
overhead (O/H) lines that limit the capability to decouple x x
Vdc
the set of equations to be solved. Controlled Controlled
To overcome these technical challenges and achieve fast voltage PWM PWM voltage
simulation of large wind power plants, three modeling source control signal control signal source
techniques have been developed and validated.
Fig. 2. Average modeling of a back-to-back PWM converter.
A. Modeling of power electronics in a wind turbine
Most modern wind turbines use either induction As a result, the average model allows quite a large
generators in a doubly-fed configuration (DFIG) or simulation time step (around 20 to 50 µs) and offers very
synchronous or permanent-magnet generators with a full- fast simulation speed. Even though harmonics are not
scale converter. These configurations are identified as types represented, this efficient modeling of converters can be
3 and 4 by [1], and are shown in Figure 1. The proposed used to conduct various power system studies in EMT and
techniques presented in this paper are applied to type 3 wind transient stability time frame.
turbines but can also be used for any type of wind turbine The second approach, the switching-functions model, has
architecture. been demonstrated and used by [3]. This model also
represents the entire back-to-back converter by two 3-phase
controlled voltage sources, but includes harmonics
generated by PWM. To do so, the voltage sources inject
switched output voltages derived from the PWM control
IG SG signals and PWM generators. Capacitor voltage variation is
Controls also considered, as in the previous case. Again, the
Controls
simulation of converters implies no switching and no
Wind turbine based on Wind turbine based on change in network topology. Figure 3 depicts the
doubly-fed induction generator synchronous or permanent magnet implementation of the switching-functions model.
(DFIG) (type 3). generator (SG, PMG) (type 4)

Fig. 1. Architecture of recent wind turbine.

At the present time, wind turbines using PWM converters

in the 1-5-kHz range are common. The simulation of PWM Capacitor voltage
Idc1 calculations Idc2
switching is very demanding for EMT simulation, since Idc1=Pac1/Vdc Idc2=Pac2/Vdc
each switching implies matrix manipulation that is very
costly in computation time. x Vdc x
Consequently, instead of using a detailed switch model, Controlled Controlled
voltage PWM PWM voltage
two different approaches were implemented. These are generator generator
source source
identified as the average model and the switching-functions
model. PWM PWM
control signal control signal
The first approach, the average model, which was used
by [2], allows the entire back-to-back converter to be Fig. 3. Switching-functions modeling of a back-to-back PWM converter.
represented by two 3-phase controlled voltage sources.
These sources are driven by the control voltages of the The time step used for this model is around 1 to 5 µs.
PWM converters. Consequently, the simulation of This technique keeps the representation of harmonics and
converters implies no switching and no change in network capacitor voltage variation. It offers precise results, as
topology. To consider the capacitor voltage (Vdc), the AC demonstrated by [3], and a fast simulation performance.
power (Pac) flowing in each converter must be equal to the This efficient modeling of power converters will satisfy
DC power (Pdc), so the following equation is applied for most large-scale power system studies, including EMT and
both converters: harmonics.
I dc = Pac / Vdc = (Vab I a − Vbc I c ) / Vdc (1)
Having the total amount of DC current flowing in or out
 B. Integration of wind turbine models into the wind increasing the L or C of the line. This artificial increase is
power plant collector network done by virtually moving an L or a C from the surrounding
Typically, a Hypersim [5] simulation is performed by power system components with negligible effect on
processing two computation steps. First, network equations simulation results.
are solved (V=ZI) to calculate the voltage at each node The first case where this technique is applicable is to
based on the current injection. Second, power system decouple the collector network from a specific U/G cable.
models, like machines, are solved to obtain the new current To achieve this, capacitances (C) of the surrounding U/G
injection at specific nodes. Thus, power system models are cables are grouped at the selected decoupling cable in order
integrated into the network as current sources, and specific to obtain a propagation delay that is greater than the time
concerns about algebraic loops and numerical delays at the step. Doing this, the global capacitance of the system is not
interface must be addressed. modified, only nearby capacitances are grouped to a
In actual implementation, these two computation steps punctual location. Figure 4 depicts this technique.
are implemented in two different simulation solvers. First,
the collector network is implemented and solved using the U/G cable used for decoupling.
Hypersim [5] real-time solver. This EMTP-like [6] Capacitances of surrounding cables
simulator uses a supercomputer to simulate the network in are grouped to increase tpropag.
parallel on many processors. Second, SimPowerSystems
(SPS) from The MathWorks Inc. [7] is used to establish the
wind turbine model itself. The C-code of this model is
generated by Matlab/RealTimeWorkshop and is exported
and simulated in Hypersim. Both solvers use discrete-time Fig 4. Decoupling the wind power plant collector network at selected U/G
cable.
modeling with a fixed time step. As a result, the SPS set of
equations is directly controlling current sources back to the
This technique is applied to U/G cables but could also be
Hypersim solver. With this approach, a joint closed-loop
applied to O/H lines. U/G cables were selected for practical
simulation is obtained, benefiting from the advantage
reasons since the C value is larger. To minimize the impact
offered by both environments. Validation has been done to
on the simulation results, the virtual displacement of
demonstrate the accuracy of this technique.
capacitances should be done in order to preserve the same
C. Decoupling of power system equations total capacitance for each positive, negative and zero-
Real-time digital simulators, based on multi-processor sequence. Using this technique, the collector network can be
system, have been relying on power system equation decoupled in numerous sub-networks, using frequency
decoupling for more than a decade [4]. Using this technique, constant distributed parameter lines (FCDP) and PI-section
the natural propagation delay of a transmission line is lines.
absorbed by the communication time between two The second case where this technique is used is to
processors. As a result, the large system impedance matrix decouple the wind turbine from the collector network
can be divided into multiple smaller matrices and can be equations. This decoupling is done at the U/G cable that
solved in parallel on many processors without numerical interconnects the turbine transformer to the collector
error. The reduced matrix size drastically diminishes the network. Again, to artificially increase the propagation
computation effort, thus improving simulation speed. delay of this U/G cable, the capacitances of surrounding
Since the total propagation time must be longer than the U/G cables are grouped, and part of the leakage inductance
simulation time step, transmission lines must be long of the transformer is included in the decoupling cable.
enough to be able to use this technique. Unfortunately, wind Figure 5 depicts this technique.
farm collector networks use short lines, so the technique
cannot be applied directly. U/G cable used for decoupling.
Therefore, in order to use equation decoupling with wind Part of the transformer leakage inductor
power plants, the propagation delay of some selected lines is used to increase tpropag.
needs to be artificially increased without affecting the
precision of the simulation results. The propagation delay IG
(tpropag) of a transmission line involves its length (l) and its
propagation speed (v), where:
1
v= (2) Fig 5. Decoupling each wind turbine at the transformer U/G cable.
LC
t propag = l (3) This technique, where part of the leakage inductance of a
v transformer is moved to the local U/G cable, is applied to a
To increase the propagation time (tpropag) of a line, two grid-side transformer connection. Since this is a delta
cases are presented based on the same approach. The connection, the zero-sequence impedance of the system is
general idea is to reduce the propagation speed by not affected.
 III. VALIDATION OF THE MODELING TECHNIQUES techniques, the same wind power plant is also modeled in
For validation purposes, a generic wind power plant SPS for comparison. Simulation results of a 3-phase fault,
model has been established. It represents typical large wind applied at the PCC of the wind power plant, are compared in
power plant from 50 to 200 MW. It is composed of 73 wind Figure 7. Results of the simulations performed using
turbines distributed into four feeders, as shown in Figure 6. Hypersim with 32 processors are compared to those on SPS
The collector network comprises mainly U/G cables and using one processor without decoupling. The first two
some O/H lines rated 34.5 kV. The power plant is graphs in this figure compare the voltage and current of
interconnected to the power system network, at the point of phase A at the PCC. The last graph compares the DC bus
common coupling (PCC), via a 34.5/230-kV transformer. voltage at wind turbine #62.
Each wind turbine uses a doubly-fed induction generator From this comparison, the effect capacitances and
(DFIG) at 575 V and 1.5-MW rating. inductances grouping is not perceptible, demonstrating the
good performance of the decoupling technique. The
simulation is performed at 20 µs in Hypersim, while using
50 µs in SPS for feasibility reasons.
To 230kV
network
#25

(kV)
3-phase
3-phases
Grounding #62
fault
transformer (kA)

PCC
(V)

34.5kV

Fig. 7. Comparison of simulation results during a 3-phase fault at the PCC

of a wind farm. A decoupled Hypersim simulation using 32 processors vs a
single-processor SPS simulation.

IV. SIMULATION IN THE EMT DOMAIN OF A120-BUS SYSTEM

1-phase NETWORK WITH TWO LARGE WIND POWER PLANTS.
fault #64
To demonstrate the performance of the proposed
techniques, a detailed EMT simulation of a 120-bus power
system network with two wind power plants (146 DFIG
wind turbines) was performed. The receiving network was a
model of the Hydro-Québec network in the Matapédia
region, as presented in Figure 8. It is composed of 120
buses, 20 transformers and 85 transmission lines. An
equivalent 3000-MW synchronous machine feeds this
power system network. Both wind power plants use the
configuration presented in Figure 6 and are interconnected
Fig. 6. The 73-turbine generic wind power plant model, based on a 1.5- to this network separated by 30 km. The simulation is
MW DFIG wind turbine and a 34.5-kV collector network.
performed on a 32-processor supercomputer using a time
Using the proposed techniques, the wind power plant step of 18.4 µs. The simulation performance is 10 times
model has been simulated in parallel on a 32-processor slower than real-time, quite fast enough for EMT and
supercomputer. Each wind turbine was decoupled from the stability studies.
collector network, and the collector network itself was Figure 9 shows the results of a 9-cycle 3-phase fault
decoupled into nine subsystems. U/G cables used for applied at the PCC (230-kV side) of the first wind farm. In
decoupling are shown in red on Figure 6. the first two graphs, the instantaneous voltage and current of
In order to validate the precision of the modeling the first wind farm, at the PCC (34.5-kV side), are shown.
The last two graphs show the superposition of active power
and DC bus voltage of two selected wind turbines. Turbines
#25 and #64 were selected for their geographic location, as
turbine #25 is close to PCC while #64 is about 8 km away.
To demonstrate that the proposed modeling can be used
to study phenomena occurring in the wind farm, simulation
results of a single-phase-to-ground fault on the collector
network of the first wind farm are also presented. Figure 10
shows the results of this unbalanced fault for six cycles.
Again, the first two graphs show the instantaneous voltage
and current of the first wind farm at the PCC, and last two,
the superposition of active power and DC bus voltage of
wind turbines #25 and #64.
Finally, to demonstrate that the proposed techniques can
also be used for stability studies, Figure 11 presents the
simulation results of a 400-MW load switch-on on this
power system network. This disturbance creates a large
frequency swing on the synchronous machine. Again, the
simulation is done in the EMT instantaneous domain, but
on-line processing of simulation results is performed to
extract fundamental positive-sequence measurements.
Grabbing results live from an ongoing simulation allows the Fig. 9. Results of an EMT simulation of a 120-bus power system network
sampling of each signal to be reduced, thereby accelerating with 2 x 73-turbine wind farms. A 3-phase fault is applied at the PCC 230-
access to understandable results. The first graph in this kV of the first wind farm.
figure shows the network frequency, resulting from the
speed of the synchronous machine. The second graph shows V. CONCLUSION
the rms voltage at the PCC of the wind farm, while the last Using the proposed modeling techniques combined with
two graphs show the active and reactive power, also at the the latest developments in supercomputer and real-time
PCC 34.5kV. simulators, it is now possible to simulate, in EMT and
transient stability, numerous wind farms on large power
Fig. 8. The 120-bus receiving network, which is a model of the actual system networks.
Hydro-Québec network in the Matapédia region, including two wind farms.

Wind Farm #1

Wind Farm #2

Synchronous
machine
 Also, the proposed modeling techniques have contributed VI. REFERENCES
to the development of a generic model of wind farm in the [1] Wind Generator Modeling Group, “Standard Wind Turbine-Generator
EMT domain. This generic model has been used as a Models,” UWIG Modeling and Interconnection User Group Meeting,
Oklahoma City, October 2006.
benchmark for other research in this area, such as the
[2] R. Gagnon, G. Sybille, S.Bernard, "Modeling and real-time simulation
development of reduced or aggregated models of a wind of a doubly-fed induction generator driver by a wind turbine," in
farm. The next development on the agenda will be Proceedings of the International Conference on Power Systems
validation of the wind farm model with actual field Transient IPST05, Montreal, June 2005, paper No. IPST05-162
[3] H. Le-Huy, G. Sybille, R. Gagnon, V.Q. Do, "Real-time simulation of
measurements. PWM power converters in a doubly-fed induction generator using
switching-function-based models," IEEE Industrial Electronics
Conference 2005, pp.1878-1883
[4] H. LeHuy, J.-C. Soumagne, "Digital real-time simulation of
transmission lines using parallel processors," IMACS-TC1 93,
Montréal, July 1993, pp. 29-32.
[5] D. Pare, G. Turmel, J.-C. Soumagne, "Validation of the Hypersim
digital real-time simulator with a large AC-DC network," in
Proceedings of the International Conference on Power Systems
Transients IPST03, New Orleans, Sept 2003, paper No. IPST03-13-5
[6] H.W. Dommel, "Digital computer solution of electromagnetic
transients in single and multiphase networks," IEEE Transactions on
Power Apparatus and Systems, vol. PAS-88, No. 4, April 1969, pp.
388-399.
[7] SimPowerSystems User’s Guide, The Mathworks Inc., 2005.
[8] A.M. Gole, A. Keri, C. Kwankpa, "Guidelines for modeling power
electronics in electric power engineering applications," IEEE
Transactions on Power Delivery, vol. 12, No. 1, January 1977, pp.
505-514.

VII. BIOGRAPHIES
Christian Larose Christian Larose received his B.Eng. degree in Electrical
Engineering in 1995 and M.Sc. degree in 1998, both from École de
Technologie Supérieure (ÉTS). He joined Institut de recherche d'Hydro-
Québec (IREQ) in 1996 as a development engineer in the Power System
Simulation Laboratory. His main interest is in the numerical real-time
simulation of power systems.

Richard Gagnon was born in Québec, Canada in 1966. He obtained his

Fig. 10. Results of an EMT simulation of a 120-bus power system network B.Sc.degree in physics engineering in 1990, his M.Sc. degree in electrical
with 2x 73 turbines wind farms. A single-phase fault is applied to the engineering in 1992 and his Ph.D. degree in electrical engineering in 1997,
collector network of the first wind farm. all from Université Laval (Québec). From 1996 to 2001, he was professor
of electrical engineering at Université du Québec à Rimouski. Since 2001,
he is a research engineer at IREQ (Hydro-Québec’s research institute). His
area of professional interests includes modeling and simulation of power
system devices and wind turbines.

Gilbert Turmel obtained his DEC in 1980 in Longueuil, Canada. He

joined the Institut de recherche d'Hydro-Québec (IREQ) in 1980. He is
working in the Power System Simulation Laboratory since 1991. He is a
senior operator of the real-time simulator. His work involved the
specification, validation and operation of the Hypersim real-time simulator
for power system study and also in giving training session on the use of the
Hypersim simulator.

Pierre Giroux obtained his BSSE degree in 1976 from the Université de
Montréal (École Polytechnique). He joined the Institut de recherche
d’Hydro-Québec in 1988 as a research engineer. His work includes design,
real-time simulation and testing of controllers for FACTS and Power
Quality Devices. He is a registered engineer with the Ordre des ingénieurs
du Québec.

Jacques Brochu (M’86-SM’06) obtained his B.A.Sc. and M.A.Sc. degrees

in electrical engineering from Université Laval in Québec City in 1981 and
1986 respectively and his Ph.D. degree from École Polytechnique de
Montréal in 1997. From 1981 to 1983, he was production engineer for
Canadian General Electric. He is currently a research engineer at IREQ
(Hydro-Québec’s research institute) where he has worked since 1985. From
1990 to 2002 he was seconded to the Centre d’Innovation sur le Transport
d’Énergie du Québec (CITEQ) in Varennes, Qc. His main areas of interest
include power electronics and power flow control devices for power
systems. He has been involved in the development of the Interphase Power
Fig. 11. Results of an EMT simulation of a 120-bus power system network Controller (IPC) technology and is the author of a reference book on the
with 2 x 73-turbine wind farms. A load switch-on creates a 2-Hz frequency subject.
swing.
Danielle McNabb received her B.Sc.A. in engineering physics in 1973, her
M.Ing. in nuclear engineering in 1980 and her M.Ing. in Electrical
Engineering in 1986, all from École Polytechnique, Université de Montréal,
Canada. She joined Hydro-Québec in 1980 where she has been involved
with control modelling and simulation for the commissioning of Gentilly 2
nuclear power plant and, since 1986, in control modeling and protection
studies for the Hydro-Québec Planning Department. She represents Hydro-
Québec TransÉnergie on the CEATI power system planning and operations
interest group and is a registered professional engineer in the province of
Québec, Canada.

Daniel Lefebvre received his B.Eng. degree in Electrical Engineering from

Ecole Polytechnique, Montreal, in 1990. Since then, he has been with
Hydro-Quebec, TransEnergie Division where he is involved in operations
planning for the Main Network. He is now a team leader in this group. He is
a registered professional engineer.