1

Doubly-Fed Induction Machine Models for

Stability Assessment of Wind Farms
Markus A. Poller

Rotor-Side

AbstractThe increasing size of wind farms requires power system stability analysis including dynamic models of the wind power
generation. Nowadays, the most widely used generator type for
units above 1MW is the doubly-fed induction machine. Doublyfed induction machines allow active and reactive power control
through a rotor-side converter, while the stator is directly connected to the grid. Detailed models for doubly-fed induction machines are well known but the efficient simulation of entire power
systems with hundreds of generators requires reduced order models. This paper presents a fundamental frequency doubly-fed induction machine model including a typical control system and discusses the accuracy of reduced order models under various operating conditions.
Index Terms doubly-fed induction machines, off-shore wind
power generation, power system stability, wind power generation,
variable speed drives

I. I NTRODUCTION

HE totally installed wind power capacity is constantly increasing.

End of June 2002, there were wind turbines with a total rated
power of 9837,27 MW installed in Germany [1]. Not only the overall
installed wind power capacity, but also the average rated power per
wind mill is constantly increasing. In Germany, during the first six
months of 2002, the average rated power per wind turbine went up to
1314 kW, which is an increase of 8% compared to the same period in
2001 [1].
Especially for wind mills above 2,0 MW, the doubly-fed induction
generator is the most widely used generator concept (e.g. GE Wind
Energy, Vestas, RE Power, Nordex, NEG Micon).
These figures clearly show that there is a strong need for power system stability analysis, including dynamic models of on- and off-shore
wind farms. For dynamic power system analysis, different models,
from fully detailed to highly reduced order models are proposed in
the literature (e.g [5],[4]), but standard doubly- fed induction machine
models for modeling large power systems are still under investigation
[2].
This paper presents an approach for standard doubly-fed induction
machine models for stability analysis. It includes models of all components, the induction generator, the rotor-side- and the grid-side converters and typical approaches for the control circuits and aerodynamics of the wind turbine.
All models have been implemented and tested in the power system
analysis program DIgSILENT PowerFactory [10].

II. T HE D OUBLY-F ED I NDUCTION M ACHINE C ONCEPT

Figure 1 shows the general concept of the doubly-fed induction generator. The mechanical power generated by the wind turbine is transformed into electrical power by an induction generator and is fed into
the main grid through the stator and the rotor windings. The rotor
winding is connected to the main grid by self commutated AC/DC
converters allowing controlling the slip ring voltage of the induction
machine in magnitude and phase angle.
Markus Poller is with DIgSILENT GmbH, Heinrich-Hertz-Str. 9, 72810 Gomaringen, Germany (email: m.poller@digsilent.de)

Grid-Side

DC/AC

Grid

Fig. 1. Doubly-fed induction generator system

In contrast to a conventional, singly-fed induction generator, the

electrical power of a doubly-fed induction machine is independent
from the speed. Therefore, it is possible to realize a variable speed
wind generator allowing adjusting the mechanical speed to the wind
speed and hence operating the turbine at the aerodynamically optimal
point for a certain wind speed range.

III. I NDUCTION G ENERATOR

is

rs

xs

rr

xr

xm

um

( ref r )

'
j
u m = u me

ir
ur

Fig. 2. Equivalent circuit of the doubly-fed induction generator

Fig. 2 shows the equivalent circuit diagram of the doubly-fed induction generator from which the model equations in a constantly, with

 rotating reference frame can be derived as follows:

 






 
 
  

   
 
 
 

  
 


 


(1)
(2)

The flux linkage can be expressed by the following equations:

  
   !#"%$ &'#"


# " ( #")!
$

(3)
(4)

The induction machine model is completed by the mechanical equation:

*  
  ")  ,+
(5)

The electrical torque is calculated from the stator current and the stator
flux:
 ,+
     $
(6)
All quantities are expressed in a stator-side per unit system.
This induction machine model of fifth order is able to represent rotor
and stator transients correctly. In stability studies however, transient
phenomena of the electrical network are usually not considered [7].
Applying the principle of neglecting stator transients to the doubly-fed
induction machine model leads to the following third order model:

     
  

 



  
  





Fig. 4. VSC-PWM converter

IV. ROTOR -

(7)
(8)

The mechanical equation is the same as in case of the fifth order model.
Neglecting rotor transients results in a first order induction machine
model that consists of steady-state stator and rotor voltage equations.

 
   
  
  



 
%




The only dynamic equation of the first order model is the mechanical
equation according to (5).

G RID -S IDE C ONVERTERS

AND

The rotor- and grid-side converters are self-commutated converters

and are usually set-up by six-pulse bridges according to Fig. 4.
Assuming an ideal DC-voltage and an ideal PWM modulation (infinite modulation frequency), the fundamental frequency line to line AC
voltage (RMS value) and the DC voltage can be related to each other
as follows:

  


(9)
(10)

U AC

U DC




"





(13)

The AC-voltage phase angle is defined by the PWM converter.

The pulse-width modulation factor  " isthe
control
of the


" variable
PWM converter. Equation (13) is valid for
. For values

larger than 1 the converter starts saturating and the level of low order
harmonics starts increasing. The complete characteristic of the PWM
converter, including the saturated range is shown in Fig. 5.

A. Rotor Current Protection

0,9

is

rs

xs

rr

xr

0,8

ir

0,7

xm

um

u 'm = u m e

rc

Fig. 3. Doubly-fed induction generator with inserted crow-bar protection ( 

and


0,6

|UAC|/UDC

us

xc

j ( ref r )

0,5
0,4
0,3
0,2
0,1

In case of faults near to the generator, rotor currents are increasing

and risk to damage the rotor-side converter. For avoiding any damages,
the rotor-side converter is bypassed when the rotor currents exceed a
certain limit (crow-bar protection, see Fig. 3 ).
While the rotor-side converter is bypassed, the generator operates
as a normal induction generator. Since the speed can be considerably
above synchronous speed before a fault occurs or the machine quickly
accelerates during a fault, the stalling point of the machine is usually
exceeded during a fault leading to very high reactive power consumption. Bypassing the rotor with an additional resistance and an additional reactance (see   and   in Fig. 3) shifts the stalling point to
a higher speed value and reduces the machines reactive power consumption considerably . This mode of operation can be considered in
the rotor-voltage (2) and the rotor flux-linkage equation (4) as follows:



 
 
'  $
  
  





#" &( #")'
! $



(11)

0
0

0,5

1,5

2,5

Pulse-Width Amplification Factor Pm

Fig. 5. PWM Converter fundamental frequency characteristic

The converter model is completed by the power conservation equation:

 !

#"%$

 
  

$

(14)

This equation assumes a loss-less converter. Because the switching

frequency of PWM converters is usually very high (typically several
hundreds Hz), switching losses are the predominant type of losses.
'
& 
Since the average switching losses are basically proportional to
,
switching losses can be considered by a resistance between the two
DC-poles in a fundamental frequency model.

(12)

The crow-bar protection is usually removed after a pre-defined time.

Additional criteria, such as voltage magnitude, can be considered for
increasing the reliability of the rotor current protection system.
In the DIgSILENT PowerFactory implementation, the crow-bar
protection is integral part of the doubly-fed induction machine model.
Criteria for inserting and removing the rotor-bypass can be flexibly
defined using PowerFactorys simulation language.

V. C ONTROL C ONCEPT
Amongst the big variety of different control concepts for variable
speed wind turbines (e.g. [4], [6]) a typical concept consisting of
dq-current regulators for the generator control and a maximum power
tracking strategy for the wind turbine control is presented and analyzed
in this paper.

A. Rotor-Side Converter
The rotor-side converter operates in a stator-flux dq-reference frame
that decomposes the rotor current into an active power (q-axis) and a
reactive power (d-axis) component.
A very fast inner control loop regulates the active- and the reactive
component of the rotor current. The current-setpoints are defined by
a slower outer control loop regulating active and reactive power (see
Fig. 6).




  

  


' 

! 
" 



 


#

'

#$% 
& 

 

E. Turbine Control

 #

&ON?P 5 QKRTS$VUXW
Y
GF IH KJML

The control concept of the grid-side converter is very similar to the

rotor-side controller concept. The grid-side converter controller operates in an AC-voltage dq-reference system. Active and reactive components of the grid-side converter currents are regulated by a fast inner
control loop (see Fig. 7). A slower outer control loop defining the qcurrent setpoint regulates the DC-voltage to a pre-defined value. The
setpoint of the d-axis component can be used for optimum reactive
power sharing between the generator and the grid-side converter or
simply kept to a constant value.

9?>@
)+*-,./0
1 2354

6
A :<= (
D
A :<=
(
A

(15)

B. Grid-Side Converter

Equation (15) shows the aerodynamic equation of a wind turbine

that relates mechanical power to wind speed and mechanical speed of
the turbine (e.g. [5]):

#

Fig. 6. Rotor-side converter controller

9;: < =

inner current control loops. On the AC side, the active power injection
is defined by the power conservation between AC and DC according to
(14). For defining the reactive power balance, the d-axis current component (reactive current) of the grid-side converter can directly be set
by a reactive power controller or a constant reactive current component
can be assumed.
A further model reduction consists of completely neglecting the
time constants of the DC voltage controller. The grid-side converter
and its controls is then modeled by a steady-state device, keeping the
DC voltage constant. With this simplification however, saturation effects (see Fig. 5)in the grid-side controller cannot be considered.

)6,./70
1 284

BC

)D , ./ 0
1 2E54

BC

with:
GF : Mechanical power of the wind turbine
H : Air density
L : Rotor radius
Q : Tip speed ratio
S : Blade pitch angle
N?P
: Power coefficient as a function Q and S
U Y : Wind speed
The tip speed ratio

Q is defined as follows:
FL
U Y

(16)

with F being the mechanical frequency of the wind turbine.

Equation (15) defines the steady-state aerodynamic behaviour of a
wind turbine; it cannot reflect dynamic stall effects correctly. An approximate method for including dynamic stall effects is described in
[9].

npXq`r
D

n$o

tTu
[X\ ] ^`_

]d
Z aXb;c Tb e

d
\^
f+g ?b hi?cjTk
s

v w

n w
l

Fig. 7. Grid-side converter controller

C. Reactive Power Control

Reactive power control is possible through the d-axis component of
the rotor- and the grid-side converters. Variable speed wind power generators can be operated at a constant power factor, which is the standard operation mode today. Alternatively, an AC-voltage controller
defining the d-axis current setpoint can be used in Fig. 6 instead of the
Q-controller, or secondary voltage control can be supported by adjusting the Q-reference value (e.g. [4]).

[Xm

vx y z

Fig. 8. Generic wind turbine model




|K}~$
?

?
{

D. Reduced Order Converter Model

For longer term simulations, when calculation time becomes an important issue, it might be desirable to neglect the very fast time constants associated with the DC intermediate circuit shown in Fig. 4.
When neglecting the DC-capacitance, the DC currents of the rotor- and
grid-side converters are forced to be equal. Consequently, the grid-side
controller must be replaced by an ideal voltage controller, without the

Fig. 9. Generic model of the pitch-control system

A generic wind turbine model for stability studies based on a maximum power tracking strategy [5] can be implemented according to
Fig. 8.

External Grid

Infeed

Line 2

In case of rotor frequencies below &"
 , active power is regulated

according to the maximum power tracking (MPT) characteristic that
defines the maximum power depending on the shaft speed as power
reference of the power controller. When the maximum shaft speed is
reached, the active power setpoint remains constant and the pitch angle
control system (see Fig. 9) starts acting driving the shaft speed back to
the maximum permitted value.
An alternative control scheme is described in [4]. Here, the speed
reference is calculated from the actually generated electrical power
(inverse MPT characteristic). As a result, the generator is driven into
optimal speed. When  "
 is reached, the pitch angle control takes
over and regulates the actual power to  "
 .

DIgSILENT

T3

Wind-Generator 3

F. Wind Fluctuations
Wind fluctuations can be modeled by varying the wind-speed-input

Connection Point

U Y of the aerodynamic block in Fig. 8.

F  "

(17)

 ")  ,+

(18)

 F  
 F   F 


F  F 
  $

(19)
(20)

In these equations turbine inertia, turbine torque and turbine frequency
F are related to the generator nominal frequency. The turbine torque
is related to the turbine power (see Fig. 8) by:

 GF

F
F

The electrical torque is defined by (6).

VI. C ASE S TUDIES

For validating the presented models, the results of simulating a
heavy and a relatively weak disturbance in the system shown in Fig.
10 are described.
The system consists of an external infeed modeled by a source
behind an impedance that is connected to a synchronous generator
through a line. And wind farm consisting of three doubly-fed induction
generators with a rating of 5MW each is connected to the mid-point of
the transmission line.
Each doubly-fed induction generator is modeled according to Fig.
11, including the three winding transformer, the induction generator,
the grid-side and the rotor-side converter and the intermediate DCcircuit.

T1

Wind-Generator 1

Wind Farm HV

T-G

When the simulated applications are limited to the impact of wind

fluctuations, it is usually sufficient to consider just a single-mass shaft
model because shaft oscillations of variable speed wind generators are
not reflected to the electrical grid due to the fast active power control.
[5].
In stability analysis however, when the system response to heavy
disturbances is analyzed, the shaft must be approximated by at least
a two mass model. One mass represents the turbine inertia, the other
mass is equivalent to the generator inertia.
The equations describing the mechanical coupling of turbine and
generator through the gear box by a two-mass model can be expressed
as follows (see e.g. [8]):

* 
F
F 
*  
 
 
 F
"

Wind-Generator 2

Gen HV Bus

G. Torsional Oscillations

T2

Wind-Farm-Cable

Line 1

Wind speed is usually modeled by superposing several deterministic

and stochastic components. However, since the response of doubly-fed
induction machines to system faults is in the center of interest of this
paper and wind speed can be assumed to be constant in these cases,
no concrete wind-speed models are described here. Generally, common wind-speed models (e.g. [9]) can be combined with the models
presented in this paper.

Gen LV Bus

G
~

Fig. 10. Power system used for model validation

A. Strong Disturbance
The first case simulates a solid three phase short circuit (duration:
200ms) at the Connection Point (see Fig. 11). The rotor-currentprotection systems bypass the rotor-side converters immediately after
the fault has been inserted (see Fig. 12 and Fig. 13). After the fault was
cleared, at t=200ms, the generators continue to operate with bypassed
converters until t=800ms, when the crow-bars are removed. When
the converters are back into operation, the system is driven back to the
initial state.
The speed shows a weakly damped torsional oscillation.
Fig. 12 compares the results obtained with a fifth order induction
generator model according to (1) and a third order model according to
(7).
In the simulation of the fifth order model, not only stator transients
of the induction machine but also the transient behaviour of transformers and lines was considered (EMT-compatible network model).
Together with the third order machine model, a quasi-steady-state
network model was used, corresponding to the classical stabilityapproach as described e.g. in [7].
When comparing the results from both models, as shown in Fig.
12, it can be noted that the reduced order model represents the average
of power and voltage very well. Higher frequency transients, as they
can be observed in the results from the detailed model, are due to network transients and are therefore not represented by the reduced order
model. The reduced order model can therefore be used for analyzing the influence of doubly-fed induction machines to the power system, but it is not possible to predict peaks in electrical power or torque
correctly. Therefore, the fifth order model together with a transient
network model is required.

Wind-Generator

DIgSILENT

DIgSILENT

5.000
2.500
0.00
-2.500
-5.000

T3WT1

-7.500
0.00

0.50
1.000
1.500
T1: Active Power in MW, 5th order model with transient network model
T1: Active Power in MW, 3rd order model with quasi-steady-state network model

..

2.000

0.00

0.50
1.000
1.500
T1: Reactive Power in Mvar, 5th order model with transient network model
T1: Reactive Power in Mvar, 3rd order model with quasi-steady-state network model

..

2.000

0.00

0.50
1.000
1.500
..
WindFarm HV: Voltage in p.u., 5th order model with transient network model
WindFarm HV: Voltage in p.u., 3rd order model with quasi-steady-state network model

2.000

0.00

0.50
1.000
1.500
G1d: Speed in p.u., 5th order model with transient network model
G1d: Speed in p.u., 3rd order model with quasi-steady-state network model

2.000

U11

L1

10.00
7.500
5.000

WT1

2.500
0.00
-2.500
-5.000

C1

PWM U1

1.600
1.200

G
~

0.80

G1d
1 x 5790 kVA

GS-G1

0.40
0.00
-0.400

Fig. 11. Wind Generator with detailed grid-side converter

1.200
1.150

The speed variable highlights differences between the two models: In the detailed model, the speed is initially reduced before the
system starts accelerating. In the reduced order model however, the
machine starts accelerating immediately after the fault was inserted.
The initial speed reduction, as shown by the detailed model, is a
consequence of decaying DC-components in the machines currents
and is known as the back-swing effect of electrical machines. Since
the third order model does not represent stator transients, the initial
back-swing is not represented by the reduced order model leading to
an immediate acceleration.
Fig. 13 compares the results of further model reductions:
1) Third order model with detailed grid-side converter, the same as
in Fig. 12
2) Third order model with simplified grid-side converter, as described in section V-D (complete reduction of the intermediate
DC-circuit.)
3) First order model with simplified grid-side converter.
All results are in good agreement. Especially the reduction of the gridside converter does not seem to have a big impact on the model accuracy when system stability is studied. The first order model however
does not represent any rotor flux transients and differs therefore from
the other curves during a short period after a heavy disturbance.
With regard to computational efficiency, some information about
the integration step-size might be of interest. All models were implemented and tested in DIgSILENT PowerFactory [10]. PowerFactory
uses a variable step size algorithm, in which the minimum step size can
be specified by the user. During the simulation, the step size is automatically increased whenever the accuracy of the numerical algorithm
allows for it.
In case of the fully detailed, fifth order machine model in combination with
a transient
network model (Fig. 12), the step size varied
 

between R ms and ms. In case of the reduced order models
together


with the steady-state network model it varied between ms and ms.
The calculation time of the third order and first order model was practically the same.

B. Weak Disturbance
In a second case, a three phase short circuit at the synchronous
generator terminal was simulated (Gen-LV-Bus). This short circuit

1.100
1.050
1.000
0.95
..

Fig. 12. Comparison of fifth order and third order model in case of a three
phase fault near to the wind generators

causes the voltage at the wind park terminal to drop to about 0.6 p.u.
(see Fig. 14). Because the rotor current protection is not triggered, the
disturbance can be classified as a fault remote from the wind park.
As Fig. 14 shows, the three wind turbines are differently loaded in
this case.
While the fault is in the system, the active and reactive power controllers remain in operation why active and reactive power is fully controlled.
Clearing the fault disturbs the system again, but active and reactive
power are well regulated. The speed of all turbines shows a weakly
damped torsional oscillation.
These simulations were carried out with the third order model and
the simplified grid-side converter model. The model represents very
well the initial active and reactive power transients and the controller
response following to the disturbance.

VII. C ONCLUSIONS
This paper presented a variable speed wind generator model suited
for stability analysis of large power systems with large on-shore and
off-shore wind farms. The presented components were the doubly-fed
induction generator, the grid-side converter, the rotor-side converter,
the aerodynamic behaviour of the wind turbine and the pitch control
system.
For simulating power fluctuations, the wind speed variable must be
fed from a measurement file, or stochastic wind models must be used
(e.g. [9]).
Possible model reductions making the model suitable for stability
assessment in large power systems were presented and discussed.

1.250
0.00

DIgSILENT

DIgSILENT

-2.000
-3.000

-1.250

-4.000

-2.500
-5.000

-3.750

-6.000

-5.000
-6.250

-7.000

0.00

0.50
1.000
1.500
T1: Active Power in MW, 3rd order model with detailed grid-side converter
T1: Active Power in MW, 3rd order model with simplified grid-side converter
T1: Active Power in MW, 1st order model with simplified grid-side converter

..

2.000

10.00

4.000

8.000

2.000

6.000

0.00

1.00
T3WT1-Generator 1: Active Power in MW
T3WT1-Generator 2: Active Power in MW
T3WT1-Generator 3: Active Power in MW

2.000

..

3.000

0.00

1.00
T3WT1-Generator 1: Reactive Power in Mvar
T3WT1-Generator 2: Reactive Power in Mvar
T3WT1-Generator 3: Reactive Power in Mvar

2.000

..

3.000

0.00

1.00
Wind Farm HV: Voltage in p.u.

2.000

..

3.000

0.00

1.00
G1d - Generator 1: Speed in p.u.
G1d - Generator 2: Speed in p.u.
G1d - Generator 3: Speed in p.u.

2.000

..

3.000

0.00

4.000
-2.000

2.000

-4.000

0.00

-6.000

-2.000
0.00

0.50
1.000
1.500
T1: Reactive Power in Mvar, 3rd order model with detailed grid-side converter
T1: Reactive Power in Mvar, 3rd order model with simplified grid-side converter
T1: Reactive Power in Mvar, 1st order model with simplified grid-side converter

..

2.000

1.250

1.100

1.000

1.000

0.75

0.90

0.50
0.80

0.25

0.70

0.00
-0.250
0.00

0.50
1.000
1.500
WindFarm HV: Voltage in p.u., 3rd order model with detailed grid-side converter
WindFarm HV: Voltage in p.u., 3rd order model with simplified grid-side converter
WindFarm HV: Voltage in p.u., 1st order model with simplified grid-side converter

..

2.000

0.60
0.50

1.200

1.030

1.150

1.000
0.97

1.100

0.94

1.050

0.91

1.000

0.88
0.85

0.95
0.00

0.50
1.000
1.500
G1d: Speed in p.u., 3rd order model with detailed grid-side converter
G1d: Speed in p.u., 3rd order model with simplified grid-side converter
G1d: Speed in p.u., 1st order model with simplified grid-side converter

..

2.000

Fig. 13. Comparison of models of various order in case of a three phase fault
near to the wind generators

The models were implemented and tested in the power system analysis package DIgSILENT PowerFactory [10]. Every reduced order
model was validated against higher order models.
The results of the test cases show that a third order induction machine model including crow bar protection together with a simplified
model of the grid-side converter provides sufficient accuracy and the
necessary computational efficiency for carrying out stability studies in
large power systems with several hundreds of machines.

R EFERENCES
[1] C. Ender, Wind Energy Use in Germany - Status 30.06.2002, DEWI
Magazin Nr. 21, August 2002
[2] S. Stapelton and George Rizopoulos, Dynamic Modelling of Modern
Wind Turbine Generators and Stability Assessment of On- and Off-Shore
Wind Farms Proceedings of the 3rd MED POWER Conference, 2002
[3] C. R. Kelber and W. Schumacher, Adjustable Speed Constant Frequency
Energy Generation with Doubly-Fed Induction Machines Proceedings
of the European Conference Variable Speed in Small Hydro, Grenoble,
France, 2000
[4] J. L. Rodriguez-Amenedo, Automatic Generation Control of a Wind
Farm With Variable Speed Wind Turbines, IEEE Transactions on Energy Conversion, Vol 17, No. 2, June 2002
[5] J. G. Slootweg, S. W. H. de Haan, H. Polinder, W. L. Kling, Aggregated
Modelling of Wind Parks with Variable Speed Wind Turbines in Power
System Dynamics Simulations, Proceedings of the 14th Power Systems
Computation Conference, Sevilla, 2002
[6] W. Hofmann and F. Okafor Doubly-Fed Full-Controlled Induction Wind
Generator for Optimal Power Utilisation, Proceedings of the PEDS01,
2001

Fig. 14. Simulation of a weak disturbance (without crow-bar insertion)

[7] P. Kundur Power System Stability and Control, McGraw-Hill, Inc.,

1994
[8] P. M. Anderson, B. L. Van Agrawal and J. E. Ness, Subsynchronous
Resonance in Power Systems, IEEE Press, New York, 1989
[9] P. Soerensen, A.D. Hansen, L. Janosi, J. Bech and B. BakJensen Simulation of interaction between wind farm and
power
system,
Technical-Report,
Risoe-R-128(EN),
2001,
http://www.risoe.dk/rispubl/VEA/veapdf/ris-r-1281.pdf
[10] DIgSILENT GmbH DIgSILENT PowerFactory V13 - User Manual,
DIgSILENT GmbH, 2002
Markus Poller was born in Stuttgart, Germany on
July 22, 1968.
In 1995, he received Dipl.-Ing. degrees from the University of Stuttgart and Ecole Nationale Superieure
des Telecommunications Paris. In 2000, he received
the Dr.-Ing. degree from the University of Hannover.
Since 1995 he works with DIgSILENT GmbH, Germany, where he is responsible for the algorithms and
models of the power system analysis software DIgSILENT PowerFactory. He is also involved in power
system studies and he presents software- and power
system analysis courses. His current research interests include wind power systems, optimal power flow dispatch and probabilistic load flow analysis.