You are on page 1of 4

Stability Enhancement of Grid-Connected DFIG with

GA Optimized PI Controller
Abstract— This work presents an efficient application of the conversion system owing to its ability to eliminate the
genetic algorithm for improving the small signal dynamic undesirable effects of parameter variations.
performance of grid-integrated (Doubly Fed Induction
Generator) DFIG system fed by wind energy. The nonlinear In this work, the design of an optimized PI controller for
differential-algebraic model of DFIG based system has been grid connected DFIG system is proposed. To this end, a
presented and linearization is performed for obtaining the small nonlinear differential-algebraic model of DFIG is presented.
signal dynamic model. Eigenvalue analysis and time domain However, the rotor-side converter is not modeled with details
based simulations are presented for visualizing the results. The as the switching frequency is far above the dynamics under
obtained result is compared with the PI controllers whose study [1]. Lastly, the genetic algorithm is applied for the
parameters are chosen based on trial-and-error method. The optimization of controller parameters to improve the small
optimized results with genetic algorithm show that the system signal stability. The subsequent sections of this paper are
shows far better performance than PI controller without GA. arranged as follows: First, the differential-algebraic model of
grid connected DFIG system is developed. The small
Keywords— Doubly-Fed Induction Generator, Genetic
disturbance stability scenario without the controller is
Algorithm, PI Controller explored next. The optimization process of PI controller gains
using the genetic algorithm along with the complete design of
I. INTRODUCTION rotor side controller of DFIG is presented next. Then, the
effect of incorporating the genetic algorithm (GA) tuned
DFIG, one of the fastest growing electromechanical device
optimized PI controller on the system transient response are
for extracting power from wind energy, had been introduced
presented through time domain simulation. Finally, the
to replace the fixed-speed induction generators. However, the
conclusions of this work are presented.
control of DFIG is far more involved than their counterpart as
they possess the ability to operate at both sub-synchronous and II. MATHEMATICAL MODELING
super-synchronous speed through proper regulation of rotor
terminal voltages [1]. Different control mechanisms such as The configuration of a variable speed wind turbine
direct torque control, field oriented control, predictive control, equipped with a DFIG is given figure 1. The varying wind
sliding mode control has been reported in the literature. energy is extracted by the turbine blade which rotates at low
speed. The gearbox is designed in such a manner that it
Traditionally, implementation of DFIG power control is converts the low-speed blade rotation into high-speed
achieved through multi-loop control scheme where the design generator shaft rotation which is suitable for power generation
is formulated with respect to stator frame [2]. On the contrary, [10]. The grid is fed by both stator and rotor electrical outputs.
direct torque control techniques are independent of coordinate However, the stator output is sent directly to the grid whereas
transformations, current control loops and precisely chosen the rotor output is sent through power converter interfaces
modulation techniques. A novel methodology, named as direct enabling the incorporation of suitable controllers.
power control strategy, was proposed in [3]. The calculation of
rotor control voltage was performed based on the knowledge A. Aerodynamic Model
of rotor position, stator flux, generated active power, reactive The amount of extracted wind power by the turbine is
power, and the resulting errors. In [4] the author presented a expressed as:
detailed comparison between field oriented control and direct Pa = 12 ρ AC p (λ , β )ν w3
torque control techniques and identified the pros and cons of (1)
each methodology. The comparisons were made based on the Performance coefficient (Cp) is further expressed as:
c7
controller’s ability in minimizing the ripples in stator current CP ( λ , β ) = c1 (
c2
− c3 β − c4 β c5 − C6 )e

λi

and electromagnetic torque and settling of transient response λi (2)


due to variation in the commanded torque. However, it was
found that due to the presence of pure integrator in the open
loop, the controllers generated noisy responses. Another direct
torque control approach was proposed in [5], where the rotor
power factor was manipulated to implement the control effort.
Predictive control was utilized in [6] for controlling the rotor-
side converter and grid-side converter of the DFIG based
study system. A simple sensor-less method was developed in
[7] for proper identification of DFIG rotor position which
eventually aided in implementing sensor-less control. Authors
in [8-9] have proposed the application of sliding mode (SMC)
based approaches in the domain of variable speed wind energy
Fig.1: Schematics of grid-integrated DFIG system

978-1-5386-3341-0/17/$31.00 ©2017 IEEE


Qtotal =
(( v
2
qs + vds2 ) − vqs2 + vds2 vb cos γ )
The coefficients c1- c7 are generally taken from standard test Xe
data set. This completes the development of the nonlinear dynamic
B. Drive Train model of the study system. The equations are linearized using
Taylor series expansion and analyzed for small signal stability.
The drive train considered in this work is assumed to have two
However, detail of the linearization is not included here for the
masses and represented as [10]:
dωt pt sake of brevity. The readers may refer to [10] for further detail
2Ht = − Tsh on the linearization process.
dt ωt (3)
1 dθ tw
= ωt − ωr
ωelB dt III. SYSTEM ANALYSIS WITHOUT CONTROLLER
(4)
dωr In this section, the small signal stability analysis of the grid-
2H g = Tsh − Te integrated DFIG system shown in fig. 1 is presented under the
dt
(5) following conditions: i) The stator voltage Vs is considered to
Where Ht: turbine inertia constant; Hg: generator inertia be constant at 1 p.u. ii) The active and reactive power injection
constant; pt: turbine input power; Tsh: shaft torque; Te: into the grid is 1 p.u. and zero respectively, and the system is
electromagnetic torque output; ωt: wind turbine angular speed. operating at ωr =1 p.u. The detail of simulation parameters of
C. DFIG Model the study system is listed in Table I.
The per unit DFIG dynamic model is expressed as [10]: TABLE I. DFIG-WT DATA
diqs ωelB  ωr 1  DFIG Data: (All data in per unit)
=  − R1iqs + ωs Ls′ids + eqs′ − ′ − vqs + K mrr vqr 
eds
dt ωs Ls′  ωs Trωs  (6) Rr 0.005 Rs 0.00706 Lm 2.9 Rr 0.00564 Ls 0.171 Lr 0.156

dids ωelB  ω 1  Drive Train Data


=  − R1ids − ωs Ls′iqs + r eds′ − eqs′ − vds + K mrr vdr  Ht 3.0 sec Hg 0.5 sec Ksh 0.3 elec. rad/ sec c 0.001 p.u.
dt ωs Ls′  ωs Trωs  (7)
d e q′ s  1  ω  
= ω e lB  R 2 i d s − e q′ s +  1 − r  e d′ s − K m rr v d r  Table II shows the eigenvalues and corresponding
dt  T ω ωs 
 r s   (8) participating states of the study system without the controller.

d e ds  1  ω   It is observed that the system is dynamically stable since none
= ω elB  − R 2 i qs − e d′ s −  1 − r  e q′ s + K m r r v q r  of the eigenvalues have positive real parts. Moreover, three
dt  T r ω s  ω s   (9)
oscillating modes are identified which are related to stator
D. Converter Model current d-q component, turbine shaft speed and twist angle,
and rotor speed and d-axis stator transient voltage. The non-
The back-to-back converter is made up of the rotor-side
oscillating mode is found to be associated with q-axis stator
converter (RSC) and the grid-side converter (GSC). The transient voltage.
intermediate dc- link capacitor stage serves as a voltage
stabilizer. The converters are equipped with IGBT based Table II: Eigenvalue analysis without controller
switches and diodes to transform dc voltage into ac voltage λ = σ + jω Dominant states
[11]. The active and reactive power balance equation for GSC λ1,2 -10.27 + j313.88 iqs, ids
is expressed as P = P + P ; Q = Q + Q λ3 -2.82 + j41.06i eds/, ωr
total s gsc total s gsc
(10)
λ5 -6.27 eqs/
,
PS = vqsiqs + vdsids QS = − vqs ids + vds iqs ,
λ6 -2.64 + j4.67i θtw, ωt
Pgsc = vqr iqr + vdr idr , Qgsc = αQtotal
Among all the modes identified in Table 2, the stator current
In this work, it is considered that there is no contribution of
mode λ1,2 is poorly damped (∼3.25%) whereas all other modes
reactive power from the converters. Hence, α = 0 , Qgsc = 0 and
are well damped (ζ >6 %).
Qtotal = QS [12].
IV. INCORPORATION OF PI CONTROLLER
E. Total Power flow equation Once we incorporate the PI controller, the state vector is
The grid is modeled by a constant voltage source (with updated as:
infinitely large inertia) in series with a reactance whose values x =  iqs ids eqs
T
′ ω r θ tw ω t xc  ; x c =  Φ id Φ iq Φ Te Φ Qs 
′ eds
T

are obtained by the application of Thevenin’s theorem [13];


where xc comprises of the state variables originating due to the
i.e. an infinite bus behind a line reactance X e . The total inclusion of PI controller in the machine-side converter. The
reactive and active powers flowing into the grid are: structures of the PI controllers considered are shown in fig. 2.
vqs2 + vds2 vb sin γ The resulting state equations are as follows:
Ptotal =
Xe (11)
analysis are presented in Table IV. Moreover, to have a
comparative analysis between GA-tuned and non-optimized PI
controller performance, the eigenvalues obtained without
optimization are also included in Table IV.

Table IV: Eigenvalue analysis with PI controllers


Without GA With GA
λ = σ + jω Dominant λ = σ + jω Dominant
states states
-470.45 iqs -7680.46 iqs
Fig. 2: PI controller structure -22.37 + j311 ids , e/ds -4725.9 ids
 = T −T
Φ -8.02 e/qs -26.77 + j307.62 e/ds, e/qs
Te e−ref e
 -4.11 Φiqr -425.05 Φiqr
Φiqr = iqr−ref − iqr
-1.2 + j9.11 θtw, ωr -6.59 + j9.52 θtw, ωr
 = Q −Q
Φ
(12) -0.03 + j0.4 Φiqs , Φidr -24.33 Φiqs
Qs s−ref s
-0.13 ΦTe -37.11 ΦTe
 =i
Φid dr−ref − idr -0.01 ωt -0.12 ωt
-211.00 Φidr
The resulting rotor voltage equations become:
vqr = Kiq KTe Ke−err + Kiq
KTe K
ΦT − Kiqiqr + iq Φiqr
From Table IV it is apparent that the GA-tuned controller
TTe e Tiq reduces the number of oscillating modes whereas increases the
(13) damping ratios of the modes. Moreover, all eigenvalues are
KQs Kid
vdr = Kid KQsQs−err + Kid ΦQs − Kid idr + Φid found to have moved further to the left half of the s-plane
TQs Tid
compared to the non-optimal PI case (without GA) which
A. Objective Function reflects larger stability margin.
The objective function (f) considered in this work is defined B. Time domain simulation
as: To validate the performance of the GA optimized PI controller
 − real ( λi )  nonlinear time domain simulation is performed. The
f = − min  
 abs ( λi )  simulation is conducted considering a torque pulse disturbance
where λi is the i-th eigenvalue of the closed loop system. In applied at 0.1 sec and removed at 0.2 sec. The dynamic
here, the aim is to minimize f in order to move all the responses of the electrical torque (Te), stator d-axis current
eigenvalues far more into the left-half of the complex s-plane, (ids), turbine speed (ωt) and rotor speed (ωr), are presented in
which will improve the small disturbance stability of the figure 3.
system under study. It is observed that the GA optimized controller shows lower
overshoot and quicker settling compared to the non-optimized
V. GA-OPTIMIZED PI CONTROLLER PERFORMANCE
PI controller.
Now, the tuning of the PI controllers is done with two 3

methods; namely- i) manually tuned trial and error based


original
optimized
2

tuning and ii) GA based tuning. Once again, for the sake of 1

brevity, detail of GA is not included in this work and can be


Te [pu]

found in the literature [14]. -1

After incorporating the PI controllers into the study system, -2

the Eigenvalue analysis is performed. The PI controller gains -3

obtained by the two methods are listed in Table III. -4


0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
time[s]

Table III: Controller parameters with and without GA


PI gains Without GA With GA 6
original
KQs 1 -1.4379 4
optimized

TQs 0.5 0.0976 2

Kid -0.01 0.3519


ids [pu]

Tid 5 0.6823
0

KTe -0.1 -0.9087 -2

TTe 0.5 0.9549 -4

Kiq -2.8 -0.1259 -6

Tiq 0.001 0.0866 -8


0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
time[s]

A. Eigenvalue Analysis with controller


To investigate the appropriateness of incorporating GA based
PI controller into the study system of fig. 1, eigenvalue
analysis is performed in MATLAB. Detail results of the
1.06
original
optimized
References
[1] Francoise Mei and Bikash Pal, “Modal Analysis of Grid-Connected
1.04

1.02 Doubly Fed Induction Generators”, IEEE Transactions on Energy


Conversion, Vol. 22, No.3, pp. 728-736, September 2007.
Wt [pu]

1
[2] R. Pena, J.C. Clare, G.M. Asher, “Doubly fed induction generator using
0.98 back-to-back PWM converters and its application to variable speed wind-
0.96
energy generation”, IEEE Proceedings on Electric Power Applications, Vol.
143, No. 3, pp. 231-241, May 1996.
0.94
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 [3] D.W. Zhi, L. Xu, “Direct power control of DFIG with constant switching
time[s] frequency and improved transient performance”, IEEE Transactions on
Energy Conversion, Volume: 22, Issue: 1, Pages: 110-118, March 2007.
1.5 [4] Domenico Casadei, Francesco Profumo, Giovanni Serra and Angelo Tani,
1.4
original
optimized
“FOC and DTC: Two Viable Schemes for Induction Motors Torque Control”,
IEEE Transactions on Power Electronics, Vol. 17, No. 5, pp. 779-787,
September 2002.
1.3
Wr [pu]

1.2
[5] Z. Liu, O. A. Mohammed and S. Liu, “A Novel Direct Torque Control
1.1 Induction Generator Used for Variable Speed Wind Power Generation”, IEEE
1 Power Engineering Society General Meeting, pp. 1-6, June 2007.
0.9
[6] Jiefeng Hu, Jianguo Zhu and D. G. Dorrell, “A Comparative Study of
Direct Power Control of AC/DC Converters for Renewable Energy
Generation”, 37thAnnual Conference on IEEE Industrial Electronics Society,
0.8
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
time[s]
pp. 3578-3583, November 2011.
[7] G. D. Marques, V. Fern˜ao Pires, S´ergio Sousa and Duarte M. Sousa, “A
DFIG Sensor-less Rotor-Position Detector Based on a Hysteresis Controller”,
VI. CONCLUSION IEEE Transactions On Energy Conversion, Vol. 26, No. 1, pp. 9-17, March
2011.
This work presents the application of GA for optimizing the PI [8] V. Utkin, J. Guldner, and J. Shi, Sliding Mode Control in Electro-
controller parameters for enhancing the small disturbance Mechanical Systems, Second Edition. CRC Press, 2 ed., May 2009.
stability of a grid connected DFIG based wind power system. [9] A. Sabanovic, L. Fridman, S. K. Spurgeon, Variable structure systems:
from principles to implementation. IET, 2004.
The performance of the designed GA based PI controllers is [10] Thomas Ackermann. (2012), Wind power on power system, Germany, A
tested against those obtained by manually tuned (non- john Wiley& sons, Ltd, publications.
optimized) controller. Eigenvalue analysis and time domain [11] B. Pokharel, "Modeling, Control and Analysis of A Doubly Fed
based simulations show that the GA tuned PI controller Induction Generator Based Wind Turbine System with Voltage Regulation,"
Master Thesis, Tennessee Technological University, December 2011.
outperforms that of the manually tuned controller [12] A. Petersson, Analysis, Modeling, and Control of Doubly-Fed Induction
performance. Generators for Wind Turbines. PhD thesis, Department of Energy and
Environment Chalmers University of Technology, 2003.
[13] C. Hamon, " Doubly-fed Induction Generator Modeling and Control in
DigiSilent Power Factory," Master Thesis, KTH School of Electrical
Engineering, 2010.
[14] R.L. Haupt, S.E. Haupt, “Practical Genetic Algorithms”, John Wiley and
Sons, New York, 2004.

You might also like