Académique Documents
Professionnel Documents
Culture Documents
HTTPS://SITES.GOOGLE.COM/SITE/JOURNALOFCOMPUTING/
WWW.JOURNALOFCOMPUTING.ORG 112
Abstract— This paper presents the design of neural network controller for Unified Power Flow Controller to reduce the oscillations of
power system network. The generalised neural network is used to minimize error signal at signified level. A generalised regression
neural network controller has been developed for conventional power system stabiliser using error signal derived from power system
network. The Power system stabilizer has been tested for different operating conditions with 250ms fault duration and carried
simulations using a single machine-infinite bus model.
Index Terms— Flexible AC transmission Systems (FACTS), Unified Power Flow Controller (UPFC), Power System Stabilizer (PSS),
Neural Network (NN), Generalized regression neural network (GRNN).
—————————— ——————————
d
0
1 INTRODUCTION dt
d
D amping oscillations in large-scale power systems are
becoming more common for stability problems. The dt
de1q
Pm Pe / M
————————————————
Dr.K.R.Sudha is with the Department of Electrical Engineering, Andhra Fig. 1 . UPFC Model
University, India.
K.Harinadha Reddy is with the Department of Electrical and Electronic
Engineering. LBR College of Engineering, India.
JOURNAL OF COMPUTING, VOLUME 3, ISSUE 3, MARCH 2011, ISSN 2151-9617
HTTPS://SITES.GOOGLE.COM/SITE/JOURNALOFCOMPUTING/
WWW.JOURNALOFCOMPUTING.ORG 113
Rotor angle
GRNN 0.6
0.4
Fig .4. Neural Network Controller
The input variable ‘x’ of the NNC must be weighted 0.2
with suitable values and added to bias. Transfer function is 0
selected as shown in above equation. Neural Network
training algorithm may be Regression Neuron model or -0.2
radial basis function model [3,19]. The output of NN -0.4
controller is also weighted with suitable value for better 0 5 10 15
performance. t
Learning algorithm Fig.6. Speed deviation at P=1.6 p.u. and at Q=0.7 p.u.
Step(1): Set input, p
Calculate vector distance between weight,W and p
P=1.6;Q=0.7
Step(2): Output of n1= (W11 - p)b1
0.015
Step(3): Output of input layer,a1 = radbis[(W11 -
p)b1] Conventional PSS
0.01
Step(4): Output of n2 = a1 W21 GRNN Controller
Step(5): Output of output layer, a2 = purelin(n2) 0.005
speed deviation
speed deviation
1 0.005
Power, Pt
0.95
0
0.9
-0.005
0.85
0.8 -0.01
0.75 -0.015
0 5 10 15
0 5 10 15
t
t
Fig.9 Terminal voltage at P=1.6 p.u. and at Q=0.7 p.u.
P=1.6;Q=0.7 Fig.12. Change in power at P=1.4 p.u. and at Q=0.6 p.u.
1.8
Conventional PSS P=1.4;Q=0.6
1.7 GRNN Controller 0.1
Conventional PSS
1.6
GRNN Controller
0.05
1.5
Votage, Vt
change in power
1.4 0
1.3
-0.05
1.2
1.1
-0.1
1
0 5 10 15
t -0.15
0 5 10 15
Now synchronous generator is assumed to a operating
t
conditions of power output P=1.4 p.u. and Q=0.6 p.u.
Transient response of power plant model is shown in Fig.13. Terminal power at P=1.4 p.u. and at Q=0.6 p.u.
figure10. Variation of speed deviation, change in power,
terminal power and voltage are shown in figures 11, 12, 13 P=1.4;Q=0.6
& 14 respectively. 1.05
Fig.10. Transient response at P=1.4 p.u. and at Q=0.6 p.u. Conventional PSS
P=1.4;Q=0.6 GRNN Controller
1.2
1
without nn controller
1 with nn controller
0.95
Power, Pt
0.8
Rotor angle
0.6 0.9
0.4
0.85
0.2
0
0.8
0 5 10 15
t
-0.2
0 5 10 15
t
JOURNAL OF COMPUTING, VOLUME 3, ISSUE 3, MARCH 2011, ISSN 2151-9617
HTTPS://SITES.GOOGLE.COM/SITE/JOURNALOFCOMPUTING/
WWW.JOURNALOFCOMPUTING.ORG 116
c hange in power
1.4 0.02
1.3
0
1.2
-0.02
1.1
1
-0.04
0 5 10 15
t -0.06
The operating level of the generator is then changed to a
0 5 10 15
low power case with P=0.4 p.u. and Q=0.2 p.u. and same
t
fault is created for 250ms duration. Performance of power
plant with and without GRNN controller has shown in Fig.18. Terminal power at P=0.4 p.u. and at Q=0.2 p.u.
figures 15,16,17,18 and 19.
Fig.15. Transient response at P=0.4 p.u. and at Q=0.2 p.u.
P=0.4;Q=0.2
P=0.4;Q=0.2 0.96
0.8 Conventional PSS
without nn controller 0.94 GRNN Controller
0.4
0.9
0.2 0.88
0.86
0
0.84
0 5 10 15
-0.2 t
0 5 10 15
Fig.16. Speed deviation at P=0.4 tp.u. and at Q=0.2 p.u. Fig.19. Terminal voltage at P=0.4 p.u. and at Q=0.2 p.u.
P=0.4;Q=0.2 P=0.4;Q=0.2
1.55
0.01
without nn controller 1.5
1.4
Votage, Vt
1.35
0
1.3
Conventional PSS
1.25 GRNN Controller
-0.005
1.2
1.15
-0.01
0 5 10 15 1.1
0 5 10 15
t t
JOURNAL OF COMPUTING, VOLUME 3, ISSUE 3, MARCH 2011, ISSN 2151-9617
HTTPS://SITES.GOOGLE.COM/SITE/JOURNALOFCOMPUTING/
WWW.JOURNALOFCOMPUTING.ORG 117
w21
dP 0.12 14 0.12 7
Pt 1.05 13 1.01 5.1 -0.5
Vt 1.6 12 1.3 6.5
P=1.4 p.u. and Q=0.6 p.u. -1
δ 1.15 13 0.59 5
dω 0.012 14 0.12 5.5 -1.5
dP 0.09 12 0.09 4.5
Pt 1.02 16 1.01 6 -2
Vt 1.55 11 1.3 6 0 5 10 15
P=0.4 p.u. and Q=0.2 p.u. t
δ 0.79 12 0.3 4.5 Fig. 22. Transient response δ21
dω 0.005 14 0.005 5.5 d1-d2
dP 0.04 15 0.04 6 0.7
Pt 0.95 14 0.92 5 without nn controller
Vt 1.5 10 1.45 6 0.6 with nn controller
0.1
G G
1 3 4 0
0.09+j0.24 0.01+j0.03 0 5 10 15
t
0.02+j0.08 Fig.23. Difference in speed of Gen.2 and Gen.3
-3 w2-w3
x 10
0.04+j0.18 0.08+j0.24
3
0.06+j0.18 Conventional PSS
5 2 GRNN Controller
0.05+j0.12
2
1
G
0
w23
-1
Transient stability and Difference in speed of Gen.2
and Gen.3 are shown in figure 23 and figure 24 -2
respectively.
-3
-4
0 5 10 15
t
JOURNAL OF COMPUTING, VOLUME 3, ISSUE 3, MARCH 2011, ISSN 2151-9617
HTTPS://SITES.GOOGLE.COM/SITE/JOURNALOFCOMPUTING/
WWW.JOURNALOFCOMPUTING.ORG 118
Fig.24. Transient response δ23 Comparison of maximum overshoot and settling time for
both Conventional PSS and GRNN Controller represented
d2-d3
0.6 in following table.
without nn controller
0.5 with nn controller Response Conventional PSS GRNN Controller
Max Settling Max Settling
0.4 Overshoot time(Sec) Overshoot time(Sec)
P=0.54 p.u. and Q=0.23 p.u.
0.3 ω12 0.0012 16 0.009 4
delta23
-0.1 5 CONCLUSION
0 5 10 15
Transient stability and Difference
t in speed of Gen.1 The new non-linear neural network controller has been
and Gen.2 is shown in figures 25 and 26. proposed for the control of UPFC. In this paper the
Fig.25 Difference in speed of Gen.3 and Gen.1 generalized regression training algorithm is used for
-3
w3-w1
training of neural network controller. This control structure
x 10
6 uses both conventional and neural network controller.
Conventional PSS Speed deviation of machine and active power of line are
GRNN Controller selected as inputs for power system stabilizer. Voltage at a
4
upfc bus is considered for training of GRNN and
oscillations of signal are damped out from light loading,
2 medium loading and heavy loading conditions.
The simulation results obtained on the single
machine infinite bus and multimachine power system with
w31
0
generalized regression neural network controller is used to
improve performance. Improvement of power system
-2 stability has been tested for fault duration of 250ms. Finally,
several fault and load disturbance results have been
-4 presented from light loading conditions to high loading
conditions.
-6
0 5 10 15 6 APPENDICES
t
6.1 Single Machine Infinite Bus
Fig.26. Transient response δ31
d3-d1 1) Generator Data
0.6 xd = 1.9, xq=1.6, xd1=0.17, τdo1=4.314s
H=4s, xe=0.3, Ke=50 τe = 0.1s
without nn controller
0.5 2) UPFC Data
with nn controller
Vdcbase=31.113Kv, MVAbase = 100, C=5500μF
0.4 Limits of UPFC data in p.u. Vcp= ±0.2, Vcr= ±0.2
3) Controllers data
0.3 Kpp = 0.3, Kip = 3, Kpq = 0.5, Kiq = 1;
delta31
-0.2
0 5 10 15
t
JOURNAL OF COMPUTING, VOLUME 3, ISSUE 3, MARCH 2011, ISSN 2151-9617
HTTPS://SITES.GOOGLE.COM/SITE/JOURNALOFCOMPUTING/
WWW.JOURNALOFCOMPUTING.ORG 119
PSS:TW=0.4,T1=0.3,T2=0.1 [12] Edris A-A. Proposed terms and definitions for flexible
Generator data: ac transmission systems (FACTS). IEEE Trans Power
Generator Delivery 1997; 2 (4):1848–53.
Ra Xd1 H
No. [13] Gyugyi L. Unified power-flow control concept for
1 0 0.20 20 flexible ac transmission system. IEEE Proc C 1992; 139
2 0 0.15 4 (4):323–31.
3 0 0.25 12 [14] Schauder CD, Hamai DM, Edris A. Operation of the
unified power flow controller (UPFC) under practical
Controllers data; constraints. IEEE Trans Power Delivery 1998;13 (2):630–9.
KPP=0.0025,Kpi=0.1085, [15] Chaudhuri B, Majumder R, Pal BC. Application of
Kpd=30×10-6(Both P and Q controller);Kpdc=0.1 multiple-model adaptive control strategy for robust
Kidc=1.0 damping of interarea oscillations in power system. IEEE
Trans Control Syst Technol 2004;12 (5):727–36.
REFERENCES [16] Wu B, Malik OP. Multivariable adaptive control of
synchronous machines in a multimachine power system.
[1] K. R. Padiyar and A. M. Kulkarni, “Control design and IEEE Trans Power Systems 2006; 21 (4):1772–81.
simulation of unified power flow controller,” IEEE Trans. [17] Narendra KS, Parthasarathy K. Identification and
on Power Delivery, vol. 13, no. 4, pp. 1348–1354, 1998. control of dynamical system using neural networks. IEEE
[2] S. Limyingcharoen, U. D. Annakkage, and N. C. Trans Neural Netw 1990;1:4–27.
Pahalawaththa, “Fuzzy logic based unified power flow [18] Amjady N. Generation adequacy assessment of power
controllers for transient stability improvement,” IEE Proc. - systems by time series and fuzzy neural network. IEEE
C, vol. 145, no. 3, pp. 225–232, 1998. Trans Power Syst 2006;21 (3):1340–1349..
[3] P.K..Dash, S.Mishra,and G.Panda “A Radial Basis [19] A.Madkiur, MA.Hossain, K.P.Dahal, H.Yu.’’Intelligent
Function Neural Network Controller for UPFC” IEEE Learning Algorithms for Active Vibration control” IEEE
Trans. on Power Systems, vol. 15, no. 4, pp. 1293–1299, Trans on Systems, Man, and Cybernetics, 2007;37 (5):1022–
November 2000. 1033.
[4] Sukumar Mishra “Neural-Network-Based Adaptive [20] D.P.Kothari.’’ Application of neural networks to power
UPFC for Improving Transient Stability Performance of systems’’ IEEE Trans Power Syst 2000; 3 (5):621–626.
Power System” IEEE Trans. on Neural Networks, vol. 17,
no.2, pp. 461–471,March 2006. Dr. K. R. Sudha received her B.E. degree in Electrical and Electronics
[5] L. Gyugyi, C. D. Schauder, S. L. Torgerson, and A. Edris, Engineering from GITAM; Andhra University 1991.She did her M.E in
“The unified power flow controller: A new approach to Power Systems 1994. She was awarded her Doctorate in Electrical
power transmission control,” IEEE Trans. on Power Engineering in 2006 by Andhra University. During 1994-2006, she
worked with GITAM Engineering College and presently she is working
Delivery, vol. 10, no. 2, pp. 1088–1097, 1995. as Professor in the Department of Electrical Engineering, Andhra
[6] M. Noroozian, L. Angquist, M. Ghandari, and G. University, Visakhapatnam, India
Anderson, “Improving power system dynamics by series-
K.Harinadha Reddy received his B.Tech degree in electrical
connected FACTS devices,” IEEE Trans. on Power Delivery,
engineering from K.U. in 1997 and M.Tech degree in Electrical Power
vol. 12, no. 4, pp. 1635–1641, 1997. Systems emphasis High Voltage Engineering from JNTUK in 2005. At
[7] K. S. Narendra and K. Parthasarathy, “Identification and present he is working as associate professor at LBR College of
control of dynamical systems using neural networks,” IEEE engineering. His research interests include HVDC transmission using
FACT controllers, AI techniques and their applications to power system
Trans. on Neural Networks, stability problems.
vol. 1, no. 1, pp. 4–27, Mar. 1990.
[8] K. G. Narendra, K. Khorasani, V. K. Sood, and R. V.
Patel, “Intelligent current controller for HVDC transmission
link,” IEEE Trans. on Power Systems, vol. 13, no. 3, pp.
1076–1083, Aug. 1998.
[9] M. Noorzian, G. Anderson, and K. Tomsovic, “Robust,
near time-optimal
control of power system oscillations with fuzzy logic, ”
IEEE Trans. on Power Delivery, vol. 11, no. 1, pp. 393–400,
1996.
[10] K.R.Sudha, K.Harinadha Reddy, “A fuzzy controller
for enhancement of Power System Stability with FACTS
device” Journal of information technology and applied
science, Vol.6,No.9, Pg33-39, Jan2010.
[11] Narain G. Hingorani, Laszlo Gyugyi “Understanding
FACTS Concepts and Technology of Flexible AC
Transmission Systems”. IEEE press, Standard Publishers
Distributors 2001:299–329.