972 IEEE TRANSACTIONS ON ENERGY CONVERSION, VOL. 24, NO.

4, DECEMBER 2009

Small-Signal Stability Analysis of a DFIG-Based

Wind Power System Under Different
Modes of Operation
Yateendra Mishra, Member, IEEE, S. Mishra, Senior Member, IEEE, Fangxing Li, Senior Member, IEEE,
Zhao Yang Dong, Senior Member, IEEE, and Ramesh C. Bansal, Senior Member, IEEE

Abstract—This paper focuses on the super/subsynchronous op- The doubly fed induction generator (DFIG) has been popu-
eration of the doubly fed induction generator (DFIG) system. The lar among various other techniques of wind power generation,
impact of a damping controller on the different modes of opera- because of its higher energy transfer capability, low investment,
tion for the DFIG-based wind generation system is investigated.
The coordinated tuning of the damping controller to enhance the and flexible control [2]. DFIG is different from the conventional
damping of the oscillatory modes using bacteria foraging technique induction generator in a way that it employs a series voltage-
is presented. The results from eigenvalue analysis are presented to source converter to feed the wound rotor. The feedback con-
elucidate the effectiveness of the tuned damping controller in the verters consist of a rotor-side converter (RSC) and a grid-side
DFIG system. The robustness issue of the damping controller is converter (GSC). The control capability of these converters give
also investigated.
DFIG an additional advantage of flexible control and stability
Index Terms—Bacteria foraging (BF), control, coordinated over other induction generators.
tuning, damping controller, Doubly fed induction generator The decoupled control of DFIG has the following controllers,
(DFIG), small-signal stability, subsynchronous operation, super-
synchronous, wind turbine (WT). namely Pref , Vsref , Vdcref , and qcref . These controllers are re-
quired to maintain maximum power tracking, stator terminal
I. INTRODUCTION voltage, dc voltage level, and GSC reactive power level respec-
tively. The coordinated tuning of these controllers by hit-and-
IND energy is one of the fastest growing industries
W worldwide. Increasing power generation from renew-
able sources, such as wind, would help in reducing carbon
trial method is a cumbersome job. The coordinated tuning using
particle swarm optimization (PSO) has been proposed [3], [4].
However, the damping of low-frequency oscillatory modes were
emissions, and hence, minimize the effect on global warming. not given due importance.
Increasing steps have been taken by the various utilities/states The impact of wind generation on the oscillatory modes is
across the world to achieve the aforementioned goal. Most of presented in [5]–[8]. The auxiliary control loop for oscillation
the states in the U.S. have Renewable Portfolios Standard (a damping that adjusts the active power command to damp the in-
state policy aiming at obtaining certain percentage of the their terarea oscillation is proposed in [5] and [6]. Moreover, a power
power form renewable energy sources by certain date) ranging system stabilizer (PSS) using a speed deviation is proposed
from 10% to 20% of total capacity by 2020 [1]. This increasing in [8]. It is reported that the presence of the PSS in the DFIG
penetration of renewable sources of energy, in particular, wind system improves the damping of the oscillations in the network.
energy conversion systems (WECS), in the conventional power Nevertheless, it is very important to optimize the controller pa-
system has put tremendous challenge to the power system oper- rameters of the PSS to achieve the best performance. However,
ators/planners, who have to ensure the reliable and secure grid the coordinated tuning of these controllers is not presented.
operation. As power generation from WECS is significantly in- Moreover, it is necessary to study the impact of these damping
creasing, it is of paramount importance to study the effect of controllers under super/subsynchronous mode of operation.
wind integrated power systems on overall system stability. In this paper, the auxiliary signal derived from ωr is added
to the rotor phase angle control to enhance the low-frequency
Manuscript received December 29, 2008; revised June 16, 2009. First damping of the system. This simple proportional–integral (PI)
published November 3, 2009; current version published November 20, 2009.
Paper no. TEC-00519-2008. controller is called damping controller. Moreover, all the DFIG
Y. Mishra is with the Midwest ISO, Carmel, IN 46082-4202 USA (e-mail: controllers, namely Pref , Vsref , Vdcref , and qcref are implemented
ymishra@midwestiso.org). in this paper. Hence, the coordinated effect of these controllers
S. Mishra is with the Indian Institute of Technology Delhi, New Delhi 110016,
India (e-mail: sukumar@ee.iitd.ac.in). on the system damping is examined. The effectiveness of damp-
F. Li is with the University of Tennessee, Knoxville, TN 37996 USA ing controller under super/subsynchronous modes of operation
(e-mail: fli6@utk.edu). is also investigated. The issue of robustness in the performance
Z. Y. Dong is with the Department of Electrical Engineering, Hong Kong Poly-
technic University, Kowloon, Hong Kong (e-mail: eezydong@polyu.edu.hk). of the damping control is discussed.
R. C. Bansal is with the University of Queensland, Brisbane, Qld. 4072, The contribution of this paper are to: 1) study the impact of
Australia (e-mail: rcbansal@ieee.org). tuned damping controller on the electromechanical modes; 2)
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org. study its impact under super/subsynchronous mode of operation;
Digital Object Identifier 10.1109/TEC.2009.2031498 and 3) propose the optimally tuned damping controller which

0885-8969/$26.00 © 2009 IEEE

MISHRA et al.: SMALL-SIGNAL STABILITY ANALYSIS OF A DFIG-BASED WIND POWER SYSTEM UNDER DIFFERENT MODES OF OPERATION 973

quadrature axes (dq axes) quantities, which form a reference

frame that rotates synchronously with the stator flux vector
[9]. The various variables are defined as: e
q s = Km r r ωs ψdr ,
e
ds = −Km r r ωs ψq r , L
s = Lss − (L2m /Lr r ), Tr = Lr r /Rr ,
2
Km r r = Lm /Lr r , ωe = ωelb ωs , R2 = Km r r Rr , and R1 =
Rs + R2 . Here, Lss is the stator self-inductance, Lr r is the
rotor self inductance, Lm is the mutual inductance between ro-
tor and stator, Rr is the rotor resistance, and Rs is the stator
resistance.
Fig. 1. DFIG system.
For balanced and unsaturated conditions, the corresponding
p.u. DFIG model can be expressed as [9]
is effective under variable operating condition. This paper is
ωs L
s diq s ωr
1

structured as follows. Section II presents the modeling of the = −R1 iq s + ωs L
s ids + eq s − e ···
DFIG system. The detailed control methodology is discussed ωe dt ωs Tr ωs ds
in Section III with special emphasis on damping controller. · · · − v q s + K m r r vq r (4)
Section IV describes the bacteria foraging (BF) algorithm for the ωs L
s dids ωr
1

optimization of the controllers parameters. Section V discusses = −R1 ids − ωs L
s iq s + e + e ···
ωe dt ωs ds Tr ωs q s
simulation and results followed by conclusions in Section VI.
· · · − vds + Km r r vdr (5)


II. MODELING OF DFIG 1 de
q s 1
ωr
= R2 ids − e + 1− e
ds − Km r r vdr
The grid-connected single machine infinite bus (SMIB) sys- ωe dt Tr ωs q s ωs
tem is considered and is shown in Fig. 1. The stator and rotor (6)
voltages of the doubly excited DFIG are supplied by the grid

1 de
ds 1
ωr
and the power converters, respectively. = −R2 iq s − e − 1− e
q s + Km r r vq r
Simulation of the realistic response of the DFIG system re- ωe dt Tr ωs ds ωs
quires the modeling of the controllers in addition to the main (7)
electrical and mechanical components. The components con-
sidered includes turbine, drive train, generator, and the back-to- where e
ds and e
q s are d- and q-axis voltages behind transient
back converter system. These components are well established reactance, respectively, ψdr and ψq r are d- and q-axis rotor
in the literature [9]; however, for the sake of completeness of fluxes, respectively, and ids and iq s are d- and q-axis stator
the paper, they are introduced in brief as shown later. currents, respectively.

A. Turbine and Drive Train C. Converter Model

The mechanical power input to the WT is considered as con- The converter model in DFIG system comprises of two
stant, i.e., the blade pitch angle do not change during the period pulsewidth modulation invertors connected back to back via
of study. In this paper, the two-mass drive train model [10] is a dc link. The RSC and the GSC act as a controlled voltage
considered and the dynamics can be expressed by the following sources. The RSC injects an ac voltage at slip frequency to the
differential equations [11]: rotor, whereas the GSC injects an ac voltage at power frequency
to the grid and maintains the dc-link voltage constant. The power
dωt
2Ht = Tm − Tsh (1) balance equation for the converter model can be written as
dt
1 dθtw Pr = Pg + Pdc (8)
= ωt − ω r (2)
ωelb dt where Pr , Pg , and Pdc are the active power at RSC, GSC, and
dωr dc link, respectively, which can be expressed as follows:
2Hg = Tsh − Te (3)
dt
Pr = vdr idr + vq r iq r (9)
where Te = Ps /ωs is the electrical torque, Tsh = Ksh θtw is the
shaft torque, Ht is the inertia constant of turbine, Hg is the Pg = vdg idg + vq g iq g (10)
inertia constant of the generator, θtw is the shaft twist angle, dvdc
ωt is the wind turbine (WT) angular speed, ωr is the generator Pdc = vdc idc = Cvdc (11)
dt
angular speed, ωs is the synchronous speed, ωelb is the electrical
where vds and vq s are d- and q-axis stator voltages, respectively,
base speed, and Ksh is the shaft stiffness.
idr and iq r are d- and q-axis rotor currents, respectively, idg and
iq g are d- and q-axis currents of the GSC, respectively, vdg and
B. Generator
vq g are d- and q-axis voltages of the GSC, respectively, vdr and
The most common way of representing DFIG for the pur- vq r are d- and q-axis rotor voltages, respectively, vdc and idc
pose of simulation and control is in terms of direct and are the voltage and current of the dc-link capacitor, respectively,
974 IEEE TRANSACTIONS ON ENERGY CONVERSION, VOL. 24, NO. 4, DECEMBER 2009

Fig. 2. Phasor diagram illustrating the operation of DFIG system [8].

and C is the capacitance of the dc capacitor. The direction of

the currents and power flow is shown in Fig 1.
The details of converter controllers are elaborated in the fol-
lowing section.

III. CONTROLLERS FOR DFIG

This section describes the controllers used for the DFIG sys-
tem. As mentioned earlier, there are two back-to-back convert-
ers; hence, we need to control these two converter sides. Pri-
Fig. 3. Control scheme for the DFIG system. (a) RSC. (b) GSC. (c) Damping
marily, these controller are known as RSC and GSC controllers. controller.
This section also introduces a new auxiliary control signal that
is added to the angle control in the RSC to enhance the damping.
This is known as damping control.
equations as follows:

A. RSC Controllers
du1
The phasor diagram in Fig. 2 describes the control scheme = Pref − P (12)
dt
[based on flux magnitude angle control (FMAC)], for the RSC
controller. u2 = Kp1 (Pref − P ) + KI 1 u1 (13)
The magnitude of the eig , internally generated voltage vector du3
= (δe i g ref + u2 − δe i g ) (14)
in the stator, depends on the magnitude of the rotor flux vector dt
ψr . This flux can be controlled by Vr , the rotor voltage. The u4 = Kp3 (Vsref − Vs ) + KI 3 u5 (15)
angle δig , between the voltage vectors eig and Vs (stator terminal
voltage and hence q-axis of the reference frame), is determined du5
= Vsref − Vs (16)
by the power output of the DFIG. Since vector eig is orthogonal dt
to ψr , the angle between d-axis and ψr is also given by δig . The du6
adjustment of the magnitude of the rotor voltage vector, |Vr | = (|eig |ref + u4 − |eig |) (17)
dt
and its phase angle, δr , is employed for the control of terminal ∆ |Vr | = Kp4 (|eig |ref + u4 − |eig |) + KI 4 u6 (18)
voltage and electrical power, respectively [8].
The configuration of the feedback controllers for the DFIG du12
= ωr ref − ωr (19)
system is shown in Fig. 3. The RSC controller is shown in dt
Fig. 3(a). One part aims at controlling the active power so as to u13 = Kp7 (ωr ref − ωr ) + KI 7 u12 (20)
track the Pref , while the second part is to maintain the terminal
voltage. ∆δr = (Kp2 (δe i g ref + u2 − δe i g ) + KI 2 u3 ) − u13 (21)
Pref is determined by the WT power speed characteristic
(Cp − λ curve) for maximum power extraction [10]. Under nor- where Kpi and KI i are the proportional and integral gain con-
mal operating condition, the active power set point Pref for the stant respectively for ith PI controller.The internal generated
RSC is defined by the maximum power tracking point, which is a
function of optimal generation speed. Mathematically, the afore- voltage vector, eig is given by, |eig | = e
2
2
ds + eq s , and the an-
 
mentioned concept can be expressed by the set of differential gle is defined as δe i g = tan−1 e
ds /e
q s .
MISHRA et al.: SMALL-SIGNAL STABILITY ANALYSIS OF A DFIG-BASED WIND POWER SYSTEM UNDER DIFFERENT MODES OF OPERATION 975

B. GSC Controllers important as the application of the damping controller in DFIG

should also ensure that the maximum power is extracted from
The GSC controller scheme is represented in Fig. 3(b). The
reference signal for the dc voltage Vdcref is set to a constant the wind.
Thus, in summary, the state equations of the DFIG are (1)–(7),
value independent of the wind speed, and Vdc is regulated by
while RSC and GSC controller state equations are (12), (14),
the following equation:
(16), (17), (22), (23), and (25). The damping controller state
dVdc 1 equation is (19). Hence, there are total 15 states of the DFIG
= (vdr idr + vq r iq r − vdg idg − vq g iq g ) (22)
dt Vdc C system including the damping controller.
du8
= Vdcref − Vdc (23)
dt IV. BF FOR THE OPTIMAL CONTROL OF DFIG SYSTEM
u9 = KP 5 (Vdcref − Vdc ) + KI 5 u8 . (24) The idea of BF is based on the fact that, natural selection tends
The reactive power set point qcref is set to zero in order to reduce to eliminate animals with poor foraging strategies and favor the
the GSC power rating. This implies that GSC only exchanges propagation of genes of those animals that have successful for-
active power with the grid, and hence, the reactive power trans- aging strategies since they are more likely to enjoy reproductive
mission to the grid by DFIG is only through the stator success. After many generations, poor foraging strategies are ei-
ther eliminated or reshaped into good ones. The Escherichia coli
du10 bacteria that are present in our intestines also undergo a forag-
= qcref − qc (25)
dt ing strategy. The control of these bacteria is basically governed
u11 = KP 6 (qcref − qc ) + KI 6 u10 . (26) by four processes, namely chemotaxis, swarming, reproduction,
elimination, and dispersal [12].
The inphase and quadrature component of the GSC voltage is
modified by
A. Chemotaxis
vg inphase = vg inphaseref + u11 xtg − (Vsref − Vs ) (27)
This process is achieved through swimming and tumbling.
vg quad = vg quadref − u9 xtg (28) Depending upon the rotation of the flagella in each bacterium
it decides whether it should move in a predefined direction
where vg inphaseref = Vs + icg quadref ∗ xtg and vg quad =
(swimming) or an altogether different direction (tumbling), in
icg inphaseref ∗ xtg , and xtg is the three-winding transformer
the entire lifetime of the bacterium. To represent a tumble, a
reactance between GSC and the stator terminal. icg inphaseref and
unit length random direction, say φ(j), is generated; this will
icg quadref are the inphase and quadrature component of GSC
be used to define the direction of movement after a tumble. In
current to the stator terminal voltage defined as icg inphaseref =
particular
Pr /Vs and icg quadref = (vds ∗ iq g − vq s ∗ idg )/Vs .
The corresponding GSC control scheme is implemented in θi (j + 1, k, l) = θi (j, k, l) + Cl(i)φ(j) (30)
this paper.
where θi (j, k, l) represents the ith bacterium at jth chemotactic
C. Damping Controller kth reproductive and lth elimination and dispersal step. Cl(i) is
the size of the step taken in the random direction specified by
Damping controller is employed in the RSC by (20), as shown
the tumble (run length unit).
in Fig. 3(c). The auxiliary signal u13 is added to the angle control
of the RSC controller to enhance the damping of low-frequency
angular oscillations. The auxiliary signal helps in increasing the B. Swarming
damping torque by controlling the angular position of the rotor During the process of reaching toward the best food location
flux vector with respect to the stator flux vector. it is always desired that the bacterium which has searched opti-
The change in the wind speed is sensed by a sensor and mum path should try to attract other bacteria so that they reach
then “ωr ref ” is determined by the maximum power tracking the desired place more rapidly. Swarming makes the bacteria
from Cp (λ, β) curve. For a constant “λ,” the ratio, ωr /Vw , is congregate into groups and hence move as concentric patterns
constant, as a result of groups with high bacterial density. Mathematically, swarming
ωr 1ref ωr 2ref ωr 1ref Vw 1 can be represented by
= ⇒ =
Vw 1 Vw 2 ωr 2ref Vw 2

S
 i 
(1 − s1 )ωs Vw 1 Jcc (θ, P (j, k, l)) = i
Jcc θ, θ (j, k, l)
⇒ =
(1 − s2 )ωs Vw 2 i=1
 

(1 − pr 1 /ps1 ) Vw 1 
S 
p
⇒ = (29) = −dattract exp −ωattract (θm −θm
i 2
)
(1 − pr 2 /ps2 ) Vw 2
i=1 m =1
where s denotes the slip, pr and ps are rotor and stator power  


S 
p
before losses. However, as the losses are very small, the values + hrep ellant exp −ωrep ellant (θm −θm
i 2
) (31)
of pr and ps can be approximately equal to Pr and Ps . This is i=1 m =1
976 IEEE TRANSACTIONS ON ENERGY CONVERSION, VOL. 24, NO. 4, DECEMBER 2009

where Jcc (θ, P (j, k, l)) is the cost function value to be added to
the actual cost function to be minimized to present a time varying
cost function. Then, Jsw = J + Jcc is calculated, where “J” is
the actual cost function to be minimized, “S” is the total number
of bacteria, “p” is the number of parameters to be optimized that
are present in each bacterium, “θm ” is the global best parameter
available during that iteration, “P (j, k, l))” is the last value
of θ stored for the next iteration, dattract , ωattract , hrep ellant ,
and ωrep ellant are different coefficients that are to be chosen
judiciously, dattract is the depth of the attractant released by the
cell and sets the magnitude of secretion of attractant by a cell,
and ωattract is the width of the attractant signal and determines
the chemical cohesion signal diffusion (smaller value makes it
diffuse more). Here, hrep ellant is the height of the repellant effect
and ωrep ellant is the measure of the width of the repellant that
controls the tendency to repel other cells.
The magnitude of dattract and hrep ellant should be same [13].
It is so that there is no penalty added to the cost function when
the bacterial population converges, i.e., Jcc of (27) will be 0.
Fig. 4. Flowchart summarizing the BF algorithm for the optimization of con-
Their numerical value should be decided based on the required troller parameters.
variation in the magnitude of the actual cost function “J” to
obtain a satisfactory result. The value of ωattract and ωrep ellant
should be such that if the euclidian distance between bacteria is V. SIMULATION AND RESULTS
large, the penalty Jcc is large. The aforementioned optimization technique is applied to a
SMIB DFIG system. The DFIG system with controllers can be
represented by the set of differential and algebraic equations
C. Reproduction (DAEs) as
The least healthy bacteria die and the all other healthier .
x = f (x, y, u)
bacteria split into two bacteria, which are placed in the same
location. This makes the population of bacteria constant. In- 0 = g(x, y, u) (32)
stead of taking the average value of all the chemotactic cost where x, y, and u are the vectors of DFIG state, algebraic and
functions, the minimum value is selected for deciding the control variables respectively. The state vector is defined by
health of the bacteria [14]. Mathematically, for particular kth
and lth, the health of the ith bacteria would be given by, x = [iq s , ids , e
q s , e
ds , ωr , θw t , . . .
i
Jhealth = min {Jsw (i, j, k, l)}.
j ε{1,2,...,N c } . . . , ωt , vdc , u1 , u3 , u5 , u6 , u8 , u10 , u12 ]T . (33)
Linearizing the previous DAE about an operating point
D. Elimination and Dispersal (x0 , y0 , u0 ) (which is obtained by the load flow at a particu-
lar wind speed), the system matrix Asys can be calculated as
It is possible that in the local environment the life of a popula- follows:
tion of bacteria changes either gradually (e.g., via consumption .
of nutrients) or suddenly due to some other influence. Events ∆ x = Asys ∆x. (34)
can occur such that all the bacteria in a region are killed or a The parameters of the DFIG system is given in Appendix.
group is dispersed into a new part of the environment. They
have the effect of possibly destroying the chemotactic progress, A. Objective Function
but they also have the effect of assisting in chemotaxis, since
dispersal may place bacteria near good food sources. From a The parameters of DFIG controllers are selected so as to
broad perspective, elimination and dispersal are parts of the minimize the following objective function:
population-level long-distance motile behavior. It helps in re- 1
ducing the behavior of stagnation (i.e. being trapped in a prema- J= (35)
(min ζi )
ture solution point or local optima) often seen in such parallel ∀i

search algorithms. This section is based on the work in [15]. The where ζi is the damping ratio of the ith eigenvalue of the system.
detailed mathematical derivations as well as theoretical aspect This objective function makes sure that the minimum damped
of this new concept are presented in [12]–[17]. eigenvalue is heavily damped and the system small-signal sta-
In this paper, optimization using BF scheme is carried out to bility is ensured. The controller parameters are optimized with
find the optimal controller parameters of the DFIG system. The the previous objective function and the performance of the al-
algorithm is presented in the flowchart, as shown in Fig. 4. gorithm is shown in Fig. 7 in the Appendix.
MISHRA et al.: SMALL-SIGNAL STABILITY ANALYSIS OF A DFIG-BASED WIND POWER SYSTEM UNDER DIFFERENT MODES OF OPERATION 977

TABLE I the application of controllers (necessary to enhance the perfor-

DIFFERENT MODES OF OPERATION OF DFIG-BASED WT SYSTEM
mance of the DFIG system) affects its damping, as shown in
Tables III–V.
Initially, the Pref , Vsref , Vdcref , and qcref are implemented
in the DFIG system. The parameters of these four controllers
are optimized using BF algorithm for a wind speed of 8 m/s
(normal mode of operation). It is observed from Table III that
mode# 3 is unstable at higher wind speed (supersynchronous
mode of operation). With the implementation of the optimized
damping controller (using BF and keeping all other controller
TABLE II
SELECTED EIGENVALUES OF THE WT SYSTEM WITHOUT ANY CONTROLLERS
parameter as constant), the system is stable. The further detailed
AT WIND SPEED OF 8 M/S discussion can be found in [18].
However, the closer look at the eigenvalues gives a new di-
mension to the whole problem. The damping controller en-
hances the damping at normal and supersynchronous mode of
operation, but at lower wind speeds (subsynchronous opera-
tion), mode# 2 becomes unstable. This mode remains unstable
even after the implementation of damping controller, as shown
in Table III. This requires further investigation as to how this
mode is excited and what can be done to make the system
stable.
B. Performance of DFIG-Based WT System Under Different The Wind speed is seldom constant and hence there is a need
Modes of Operation for a robust damping controller which can perform well under
various operating conditions. Therefore, to further investigate
It is important to emphasize the steady-state operation of the
the impact of damping controller at different operating condi-
DFIG-based WT system under different modes of operation.
tions, the DFIG system is studied for supersynchronous and
Based on the average wind speed in the particular area, the WT
subsynchronous modes of operation.
(blade radius) can be designed to operate DFIG near the syn-
chronous speed. The near synchronous speed would be desired
by the manufacturer so as to extract maximum power from the D. Optimal Tuning of DFIG Controllers
stator and hence put less burden on rotor convertors. This would
The robust performance of DFIG controllers is desired at
help the rotor current to not exceed RSC and GSC thermal rating
all modes (normal and super/subsynchronous) to ensure the
in the event of wind speed variation. Hence, this would enhance
stable operation of the WT under stochastically varying wind
the life cycle of the convertors. From the operation point of view,
speed. Hence, it is necessary to find the optimal parameters of
the additional rotor power at increased wind speed would give
all the controllers (including damping controller) for the stable
some reserve which can be added to the grid.
operation under changing wind speeds. Usually, the wind speed
With the given WT rating (in the Appendix), DFIG operates
varies in the range of 6–14 m/s. However, these extreme wind
at supersynchronous mode when the wind speed (Vw ) is more
speeds are rare and hence the speed of 8 m/s is selected for
than 8 m/s and at subsynchronous mode at lower wind speeds.
the near synchronous or normal mode of operation (it depends
Table I shows different modes of operation of the given DFIG
on turbine manufacturer and the average wind speed selection).
WT system. At supersynchronous mode, the power is supplied
For the optimal performance, controllers should be optimized at
by stator as well as rotor, whereas rotor absorbs power at sub-
the speed near to the rated/normal speed. Therefore, the DFIG
synchronous mode of operation. The power coming in to the
controllers are optimized at three different wind speeds (Vw ),
RSC and going out of GSC is considered positive, as shown in
i.e., 8 m/s (normal), 8.5 m/s (supersynchronous), and 7 m/s
Fig 1.
(subsynchronous). This is shown in Tables III–V.
When the controllers are optimized for Vw = 8 m/s, the sys-
C. Need of Damping Controller
tem is stable for normal (Vw = 8 m/s) and supersynchronous
The eigenvalues of the WT system without any control at (Vw > 8 m/s) mode of operation. However, at subsynchronous
wind speed of 8 m/s is shown in Table II. The system looks (Vw < 8 m/s) mode of operation, mode# 2 becomes unstable, as
stable with well damped eigenvalues. The first mode is stator shown in Table III. It is interesting to observe that the damping
or electrical mode and the second is electromechanical mode, controller still works well at this mode of operation, as mode#
which can be identified by looking at the participation factors. 3 is stable with damping controller. However, mode# 2 has be-
As electrical state (eds ) and mechanical state (ωr ) participates in come unstable because the participation of the mechanical state
the second mode, hence this mode is electromechanical mode. (ωr ) has been reduced and that of electrical states (e
ds , e
q s )
The stator mode has the lowest damping ratio but its frequency have increased at the subsynchronous mode of operation, which
is high and hence out of the range of interest. The low-frequency can be seen from Table III. This suggests that the system is not
mode, i.e mechanical mode (0.53 Hz) is well damped. However, small-signal stable at subsynchronous mode of operation, if the
978 IEEE TRANSACTIONS ON ENERGY CONVERSION, VOL. 24, NO. 4, DECEMBER 2009

TABLE III
SELECTED EIGENVALUES OF THE WT SYSTEM

TABLE IV
SELECTED EIGENVALUES OF THE WT SYSTEM

controllers are optimized at near synchronous (normal) wind operating wind speeds. To further verify this, the controllers are
speed. optimized at Vw = 7.5 m/s and Vw = 6.5 m/s. Table VI summa-
In the next step, the controller are optimized at wind speed rizes the variation of mode# 2 when controllers are optimized
(Vw ) of 8.5 m/s. Again the system is unstable at subsynchronous at different wind speeds. It is observed that all the modes are
mode of operation. This can be illustrated from Table IV. As stable if controllers are optimized at any subsynchronous speed.
shown in Table IV, the optimization at supersynchronous speed Again, the speed depends on the rating of the DFIG WT. For
would help in eliminating the electromechanical mode (mode# example, if the installed WT has near synchronous operation at
3). But on the downside, the system has an additional mode, Vw = 8 m/s, then the controllers can be designed at any sub-
mode# 4, associated with state Vdc , voltage across the dc-link ca- synchronous speed (Vw < 8 m/s). This would ensure the stable
pacitor. This can have detrimental effect on the system stability performance of DFIG system across the wide range of wind
if not properly damped. Moreover, it affects the convertor rating speed from subsynchronous (Vw < 8 m/s) to supersynchronous
and therefore the oscillations of this kind should be damped (Vw > 8 m/s) mode of operation.
if the controllers are optimized for this operating condition.
Mode# 2 is still unstable at lower wind speeds (subsynchronous
operation). E. Super/Subsynchronous Mode of Operation
Lastly, the controllers are optimized at lower wind speeds, i.e., It is interesting to observe the change in the frequency of os-
Vw = 7 m/s, and the eigenvalues of the system with the different cillations of different eigenmodes under super/subsynchronous
controllers are shown in Table V. The system is stable for all the operation of DFIG. The eigenvalues of the DFIG-based WT
MISHRA et al.: SMALL-SIGNAL STABILITY ANALYSIS OF A DFIG-BASED WIND POWER SYSTEM UNDER DIFFERENT MODES OF OPERATION 979

TABLE V
SELECTED EIGENVALUES OF THE WT SYSTEM

TABLE VI
BEHAVIOR OF MODE# 2 WHEN THE CONTROLLERS ARE OPTIMIZED AT DIFFERENT WIND SPEEDS

system is observed under super/subsynchronous mode of op- F. Nonlinear Simulations

eration. It is observed that the frequency of mode# 2 has The nonlinear simulation of the DFIG system is observed to
increased from ∼10 Hz (normal operation) to ∼40 Hz (su-
validate the efficacy of the damping controller and its perfor-
per/subsynchronous operation) when the controllers are opti- mance under variable wind speeds. The optimized controller
mized at 8 m/s. Mode#2 has become electrical mode from elec- parameters at different wind speeds are used and the system
tromechanical mode. This can be easily verified by the partic-
is simulated in MATLAB using ODE15s. The efficacy of the
ipation factor analysis, as shown in Table III. Moreover, when optimization procedure is observed.
controllers are optimized for the wind speed of either 8.5 or The system is subjected to a small perturbation by a small
7 m/s, the frequency of mode# 2 changes to ∼20 Hz for su-
change in the wind speed at 1.0 sec. It amounts to 5% decrease
per/subsynchronous operation, as shown in Tables IV and V. in electrical torque. The system torque returns to its original
Therefore, it is still electromechanical mode but have more par-
value after 200 ms. The response of the DFIG system with and
ticipation of electrical states (e
ds , e
q s ) than mechanical state
without damping controller is shown for supersynchronous and
(ωr ). The damping of mode# 2 is observed when the controllers subsynchronous mode of operation in Figs. 5 and 6, respec-
are optimized at different wind speeds, as shown in Table VI. It is
tively. The controller parameters are optimized at Vw = 7 m/s.
observed that damping is negative under subsynchronous mode
As expected from the eigenvalue analysis, without damping
of operation (wind speed is 6 m/s) when controllers are opti- controller, the system is unstable for supersynchronous mode of
mized at 8.5 and 8 m/s. However, when optimized at 7.5, 7, and
operation (wind speed is 10 m/s) because of negative damping
6.5 m/s, mode# 2 is stable. This confirms that the system stability of mode#3.The system is very well behaved if the optimized
can be ensured at all wind speeds if the controllers are optimized damping controller is used. Under subsynchronous mode of
at any subsynchronous speed.
operation (wind speed is 6 m/s), the system is stable with and
980 IEEE TRANSACTIONS ON ENERGY CONVERSION, VOL. 24, NO. 4, DECEMBER 2009

Fig. 5. Dynamic response of the DFIG system when the wind speed is 10 m/s (supersynchronous) and the controllers are optimized at 7 m/s.

Fig. 6. Dynamic response of the DFIG system when the wind speed is 6 m/s (subsynchronous) and the controllers are optimized at 7 m/s.
MISHRA et al.: SMALL-SIGNAL STABILITY ANALYSIS OF A DFIG-BASED WIND POWER SYSTEM UNDER DIFFERENT MODES OF OPERATION 981

extended to multimachine DFIG-based WT system. Neverthe-

less, this paper does provide a good initial study of the DFIG
system with controller. Computations with multimachine DFIG
system will be required to confirm the obtained results and de-
termine if its possible to quantify the impact of DFIGs on power
system stability.

APPENDIX
A. Parameters of the SMIB DFIG System (p.u.)
Ht = 4; Hg = 0.4; Xm = 4; Lm = 4; xtg = 0.55; C = 0.01;
xe = 0.06; Lss = 4.04; Lr r = 4.0602; Rs = (Xm /800); Rr =
1.1*Rs .

Fig. 7. Performance of BF algorithm for V w =7.5 m/s. B. Parameters Used for the Optimization (BF Algorithm)
S = 4; Cl = 0.07; dattract = 1.9; ωattract = 0.1; hrep ellant =
TABLE VII dattract ; ωrep ellant = 10.
OPTIMIZED CONTROLLER PARAMETERS FOR DIFFERENT WIND SPEEDS
The earlier values of BF algorithm parameters are used for
optimizing the controller parameters. First, all the six controller
parameters (i.e., six pairs of KP m and KI m ) are optimized and
then the damping controller (KP 7 and KI 7 ) is optimized for a
given wind speed. This process is repeated for different wind
speeds. Fig 7 shows the performance of the algorithm for a
particular wind speed. Table VII shows the optimized controller
parameters using BF algorithm for different wind speeds.

REFERENCES
[1] U.S. Dept. of Energy. (2007, Jun.). EERE state activities and part-
nerships. [Online]. Available: http://apps1.eere.energy.gov/states/maps/
renewable_portfolio_states.cfm
without damping controller. The active and reactive power avail- [2] P. B. Eriksen, T. Ackermann, H. Abildgaard, P. Smith, W. Winter, and
able at the GSC is also presented. It can be seen that the GSC J. M. R. Garcia, “System operation with high wind penetration,” IEEE
reactive power variation is well damped. Similar consistent ob- Power Energy Manage., vol. 3, no. 6, pp. 65–74, Nov. 2005.
[3] W. Qiao, G. K. Venayagamoorthy, and R. G. Harley, “Design of optimal
servations are made for other set of wind speeds and optimized PI controllers for doubly fed induction generators driven by wind turbines
controller parameters. using particle swarm optimization,” in Proc. Int. Joint Conf. Neural Netw.,
Canada, Jul. 2006, pp. 1982–1987.
[4] F. Wu, X. P. Zhang, K. Godfrey, and P. Ju, “Small signal stability analysis
VI. CONCLUSION and optimal control of a wind turbine with doubly fed induction generator,”
IET Gener. Transm. Distrib., vol. 1, no. 5, pp. 751–760, 2007.
The damping controller gives promising results in damping [5] Z. Miao, L. Fan, D. Osborn, and S. Yuvarajan, “Control of DFIG based
out the low-frequency oscillations and hence improves the sys- wind generation to improve inter area oscillation damping,” in Proc. IEEE
PES General Meeting, Pittsburgh, PA, Jul. 2008, pp. 1–7.
tem stability of the grid-connected DFIG system. The tuning [6] L. Fan, Z. Miao, and D. Osborn, “Impact of doubly fed wind turbine
of the controllers is emphasized to ensure the stable operation generation on inter-area oscillation damping,” in Proc. IEEE PES General
under variable wind speed. The supersynchronous, normal, and Meeting, Pittsburgh, PA, Jul. 2008, pp. 1–8.
[7] D. J. Vowles, C. Samasinghe, M. J. Gibbard, and G. Ancell, “Effect of
subsynchronous mode of operation of DFIG-based WT is thor- wind generation on small signal stability- a New Zealand example,” in
oughly investigated. It is observed that when the tuning of the Proc. IEEE PES General Meeting, Pittsburgh, PA, Jul. 2008, pp. 1–8.
controllers is done at any subsynchronous speed, the system is [8] F. M. Hughes, O. A. Lara, N. Jenkins, and G. Strbac, “A power system
stabilizer for DFIG-based wind generation,” IEEE Trans. Power Syst.,
stable for all modes of operations. The change in the frequency vol. 21, no. 2, pp. 763–772, May 2006.
of the electromechanical modes under super/subsynchronous [9] F. Mei and B. C. Pal, “Modeling snd small signal analysis of a grid
operation is presented. This study would help in understanding connected doubly fed induction generator,” in Proc. IEEE PES General
Meeting, San Francisco, CA, 2005, pp. 358–367.
the interaction of the oscillatory modes of DFIG-based WT with [10] S. K. Salman and A. L. J. Teo, “Windmill modeling consideration and
other components of power systems. Further, it would also help factors influencing the stability of a grid-connected wind power based
in proper tuning of the DFIG controllers to enhance the system embedded generator,” IEEE Trans. Power Syst., vol. 18, no. 2, pp. 793–
802, May 2003.
small-signal stability. [11] F. Mei and B. C. Pal, “Modal analysis of grid connected doubly fed induc-
With the increasing penetration of DFIG-based wind farms tion generator,” IEEE Trans. Energy Convers., vol. 22, no. 3, pp. 728–736,
into the grid, it is important to study the implications of large Sep. 2007.
[12] K. M. Passino, “Biomimicry of bacterial foraging for distributed optimiza-
scale DFIG systems on grid stability. As this study is based on tion and control,” IEEE Control Syst. Mag., vol. 22, no. 3, pp. 52–67, Jun.
SMIB DFIG system, conclusions of this paper should not be 2002.
982 IEEE TRANSACTIONS ON ENERGY CONVERSION, VOL. 24, NO. 4, DECEMBER 2009

[13] S. Mishra, “Hybrid least-square adaptive bacteria foraging strategy for Fangxing (Fran) Li (S’98–M’01–SM’05) received
harmonic estimation,” IEE Proc.-Gener. Transm. Distrib., vol. 152, no. 3, the B.S.E.E. and M.S.E.E. degrees from Southeast
pp. 379–389, May 2005. University, Nanjing, China, in 1994 and 1997, re-
[14] M. Tripathy and S. Mishra, “Bacteria foraging-based solution to optimize spectively, and the Ph.D. degree from Virginia Tech,
both real power loss and voltage stability limit,” IEEE Trans. Power Syst., Blacksburg, in 2001.
vol. 22, no. 1, pp. 240–248, Feb. 2007. He has been an Assistant Professor at the Univer-
[15] S. Mishra, “A hybrid least square-fuzzy bacteria foraging strategy for sity of Tennessee (UT), Knoxville, and an Adjunct
harmonic estimation,” IEEE Trans. Evol. Comput., vol. 9, no. 1, pp. 61– Researcher at Oak Ridge National Laboratory, since
73, Feb. 2005. August 2005. He was with ABB, Raleigh, NC, as
[16] M. Hunjan and G. K. Venayagamoorthy, “Adaptive power system stabiliz- a Senior and then a Principal Engineer for four and
ers using artificial immune system,” in Proc. IEEE Symp. Artif. Life, Apr. a half years. His current research interests include
2007, pp. 440–447. wind power integration, energy market, reactive power, and distributed energy
[17] S. Mishra, M. Tripathy, and J. Nanda, “Multimachine power system sta- resources.
bilizer design by rule based bacteria foraging,” Electr. Power Syst. Res., Dr. Li is a registered Professional Engineer in North Carolina.
vol. 77, no. 12, pp. 1595–1607, Oct. 2006.
[18] Y. Mishra, S. Mishra, M. Tripathy, N. Senroy, and Z. Y. Dong, “Improving
stability of a DFIG-based wind power system with tuned damping con-
troller,” IEEE Trans. Energy Convers., vo. 24, no. 3, pp. 650–660, Sep.
2009.
Z. Y. Dong (M’99–SM’06) received the Ph.D. de-
gree from the University of Sydney, Sydney, N.S.W.,
Australia, in 1999.
He was with the University of Queensland, Aus-
tralia, and the National University of Singapore. He
was also with Powerlink Queensland, and Transend
Networks, Tasmania, Australia, both in power sys-
tem planning area. He is currently with Hong Kong
Polytechnic University, Kowloon, Hong Kong. His
Yateendra Mishra (M’09) received the Ph.D. degree current research interests include power system plan-
from the University of Queensland, Brisbane, Qld., ning, power system security, stability, and control,
Australia, in 2009, the B.E. degree from BIT Mesra, load modeling, electricity market, and computational intelligence and its appli-
Mesra, Ranchi, in 2003, and the M.Tech. degree from cation in power engineering.
Indian Institute of Technology Delhi, New Delhi, In-
dia, in 2005.
For six months, he was a Visiting Scholar in the
University of Tennessee, Knoxville. He is currently a
Transmission Planning Engineer with Midwest ISO,
Carmel, IN. His current research interests include Ramesh C. Bansal (S’99–SM’03) received the M.E.
distributed generation, doubly fed induction gener- degree from Delhi College of Engineering, New
ator (DFIG) based wind generation systems, and power system stability and Delhi, India, in 1996, the M.B.A. degree from In-
control. dira Gandhi National Open University, New Delhi, in
1997, and the Ph.D. degree from the Indian Institute
of Technology Delhi, New Delhi, in 2003.
During June 1999–December 2005, he was an
Assistant Professor in the Department of Electrical
and Electronics Engineering, Birla Institute of Tech-
nology and Science, Pilani, India. During February
2006–June 2008, he was with the School of Electri-
S. Mishra (M’97–SM’04) received the B.E. degree cal and Electronics Engineering Division, School of Engineering and Physics,
from the University College of Engineering, Burla, University of the South Pacific, Suva, Fiji. He was with the Civil Construction
India, in 1990, and the M.E. and Ph.D. degrees Wing, All India Radio, for nine years during August 1989– August 1998. He is
from the Regional Engineering College, Rourkela, currently a Faculty member in the School of Information Technology and Elec-
India, in 1992 and 2000, respectively. trical Engineering, the University of Queensland, Brisbane, Qld., Australia. He
In 1992, he joined the Department of Electrical has authored or coauthored more than 110 papers in national/international jour-
Engineering, University College of Engineering, as a nals and conference proceedings. He is a member of the Editorial Board of the
Lecturer, and subsequently became a Reader in 2001. Institution of Engineering and Technology (IET), Renewable Power Generation,
He is currently an Associate Professor in the Depart- Electric Power Components and Systems Energy Sources, Part B: Economics,
ment of Electrical Engineering, Indian Institute of Planning and Policy. His current research interests include reactive power con-
Technology, New Delhi, India. His current research trol in renewable energy systems and conventional power systems, power system
interests include fuzzy logic and artificial neural network applications to power optimization, analysis of induction generators, and artificial intelligence tech-
system control and power quality. niques applications in power systems. Dr. Bansal is an Editor of the IEEE
Dr. Mishra is a Fellow of the Indian National Academy of Engineering TRANSACTIONS OF ENERGY CONVERSION AND POWER ENGINEERING LETTERS,
(INAE). He has been honored with many prestigious awards such as the Indian an Associate Editor of the IEEE TRANSACTIONS OF INDUSTRIAL ELECTRONICS.
National Science Academy (INSA) Young Scientist Medal-2002, the INAE He is a member of the Board of Directors of the International Energy Foundation
Young Engineer’s Award-2002, the recognition as DST Young Scientist 2001– (IEF), AB, Canada. He is also a member of the Institution of Engineers (India)
2002, etc. and a Life Member of the Indian Society of Technical Education.