Online Control of SVC Using ANN Based Pole Placement Approach

Download as pdf or txt
Download as pdf or txt
You are on page 1of 5

PRZEGLD ELEKTROTECHNICZNY (Electrical Review), ISSN 0033-2097, R.

88 NR 7a/2012 33
Ercan KSE
1
, Kadir ABACI
2
, Saadettin AKSOY
3

University of Mersin (1-2), University of Sakarya,.Turkey (3)


Online Control of SVC Using ANN Based Pole Placement
Approach


Abstract. In this study, an online Artificial Neural Networks based Pole Placement algorithm is proposed to stabilize the 2-bus system with nonlinear
SVC and variable power. It uses pole placement method. Proposed algorithm are examined in point of improving the bus voltages which change
under different demanding powers and the temporary state performance are compared with PID control based results. Simulation results
demonstrate a good performance and robustness.

Streszczenie. W artykule zaprezentowano online algorytm Pole Placement z wykorzystaniem sieci neuronowych do stabilizacji dwu-szynowego
systemu z nieliniowym, SVC (Static Var Compensator) i zmienn moc. (Sterowanie SVC przy wykorzystaniu algorytmu Poler Placement i sieci
neuronowych)

Keywords: SVC, PID Control, Online ANN based Pole Placement, Linearization, Acermann Method.
Sowa kluczowe: SVC static var compensator, sieci neuronowe.


Introduction
It is well known that shunt and series compensation can
be used to improve the power system stability. With the
improvement in current and voltage handling capabilities of
power electronic devices that have allowed for the
development of Flexible AC Transmission Systems
(FACTS), the possibilty has been arise of using different
types of controllers for efficient shunt and series
compensation [1]. The most popular type of FACTS devices
in terms of application is the SVC. These devices are well
known to improve power system properties such as steady-
state stability limits, voltage regulation and var
compensation, dynamic over-voltage and under voltage
control, and damp power system oscillations [2, 3].
Linearization is one of the most important issues for
control of nonlinear systems. There are lots of study in the
literature regarding linearization. For example, Taylors
series based expansion linearization method is applied to
several examples, i.e. the asynchronous slip-ring motor and
non-linear diode [4].In an another study, the design
principles are examined using the direct feedback
linearization technique to desing a nonlinear coordinated
controller for the power systems [5, 6].
Today, the PID controllers, are widely used for
automatic control mechanisms, and various methods have
been developed for automatic parameter tuning of the PID
[7, 8]. FACTS devices usually include PID controllers [9,
10]. On the other hand, in the modern control pole
placement-based controller design techniques have been
widely used. For example, Arvantis has presented some
new approaches in adaptive pole placement problem for
linear systems [11]. In another study, active vibration control
applications have been tested for the applicability of the
pole placement-based design techniques [12]. In Shakirs
study is examined the performance of a state feedback
controller method used in the pole placement method [13].
An artificial neural network (ANN) is a self-adapting
method which has developed rapidly all over the world
since the 1980s and is extensively used in many fields,
such as image, speech and voice recognition, complex
computations as well as trend prediction [14,15,16].
Moreover, It is being used successfully in many areas of
power systems,such as power system control, prediction
etc. [17, 18, 19]. For example, Girish has disscussed
interesting applications of ANNs; load forecasting, dynamic
safety evaluation and diagnostics [20].
In this study, a pole placement method is proposed for
online control of SVC. A stable point is selected for pole
placement. The system which is nonlinear is linearized
around this point, and required feedback vector is obtained
from ANN. Voltage stability of SVC is analyzed using
proposed algorithm and compared to PID controller replies.
Simulation results show that the output voltage is stable for
reference voltage and various demand power. Furthermore
PV curves show the system is stabilized in very short time.

System Model
Plant and SVC model
We consider a model for the generator and an SVC
system that can be described by a two-bus system, shown
in Fig. 1. This model can be viewed as a generalized case
where the SVC is located at the end-point of a transmission
line.


Fig. 1 The two-bus SVC system

Generator model
The system model for excitation control design and
stability analysis is usually that of a single generator infinite-
bus system. where (t), power angle of the generator; (t),
rotor speed of the generator; Pm, mechanical input power;
P
e
(t), active electrical power delivered by the generator; D
G
,
damping constant; M, inertia constant; X, reactance of the
transmission line constant. The model can be written as
follows,

(1)
1 2
m G
2 1 2
2 2 S d
( ) ( )
sin 1
( ) ( )
cos 1 1
( ( ) )
t t
VV
t P D
M X
VV
V v B kP
X X
o e
o
e e
o
t
=
=
= +


where
1 G
1, X=0.5, 1, =8, k=0.25, D =0.1. M V t = =


Control
PWM Signals
B
c

B
L

G G
P jQ +

d d
P jQ +

R jX +
1 1
V o Z

B
s

2 2
V o Z

34 PRZEGLD ELEKTROTECHNICZNY (Electrical Review), ISSN 0033-2097, R. 88 NR 7a/2012
The steady state load demand is modeled through the
parameter P
d
, under the assumption that reactive power
load demand is directly proportional to the active power
demand, i.e, Q
d
=k.P
d
; this parameter is used here to carry
out the voltage collapse studies. SVC operated capacitive
mode figures out compensation effect for power system
stability. To simplify the stability analysis, resistance and
line susseptance are neglected (R=0, B
L
=0), P
m
=P
d
. The
value belonging to limit point of system is
max
d
0.78078 P =
pu
[21]. The system acts non-linear at the value P
d
> 0.78078
pu.

SVC model
The SVC systems are used with an electrical
transmission line connecting various generators and loads
at its sending and receiving end. It consists of a fixed
capacitor in shunt with a thyristor-controlled inductor. The
susceptance of the inductor is controllable through
controlling the firing angle of the thyristor with PWM (pulse
width modulation) control signals.
The SVC model equations is given as follows:

(2)
S
L
(2 sin(2 ) 2 )
2 sin(2 ) 2
L
S
B
B
B
B
t o o
t
t
t o o
+
=
=
+

where B
L
(t) the susceptance of the inductor in SVC; the
firing angle; B
S
(t) the full susceptance of the SVC.

Taylor Series Expansion Based Linearization
Taylor series expansion is widely used for linearization
of the non-linear systems. This method can be applied to
the power system shown in Fig. 1.
Let f(x) be a one-variable function. Taylor series
expansion of the f(t) at point
x
can be written as

(3) ( ) ( ) ( ) higher-deg. terms
x x
df
f x f x x x
dx
=
= + +
is obtained. A function is expressed in a great ratio by the
first terms of the expansion. Therefore, higher degree terms
can be ignored.
We consider following non-linear state sapace model:

(4.a)
(t) = ( , ) x f x u

(4.b)
(t) = ( , ) y g x u


Its Taylor series based linearized state space model
can be written as follows:
(5.a) x(t) = Ax(t) + Bu(t)
(5.b) y(t) = Cx(t) + Du(t)

Where,
(6)
1 1 1 1 1
1 2 3 1 2
2 2 2 2 2
1 2 3 1 2
3 3 3 3 3
1 2 1 2 3
1 1 1 1 1
1 2 3 1 2
, ,
,
f f f f f
x x x u u
f f f f f
x x x u u
f f f f f
u u x x x
g g g g g
X X X u u
( ( c c c c c
( (
c c c c c
( (
( (
c c c c c
= = ( (
c c c c c
( (
( (
c c c c c
( (
c c c c c ( (

( ( c c c c c
= =
( (
c c c c c

A B
C D

From Eqs. (1), the generator model is highly nonlinear.
The systems of state variables (t), (t), v
2
,and control
variable B
S
is described. In state space modeling A, B, C
and D represent the case of Taylor series expansion around
equilibrium point with the transient components.
(7)
1 2 3 2 d 1 S 2
2
3 1 2 1 2
2
3 1 3
2
3 2 1 3
2 3 1
, , , ,
2 sin 0.1
0.25 cos 0.25
0.125 0.03125
(t)
x x x v P u B u
x X f
x x x x u f
x x x x
x u u f
y v x g
o e
e
= = = = =
= = =
= + =
=
+ =
= = =
1
2
3


The state variable values of operation point,
(8)
2 2
0.36, 0.8465, 0.6, 0, 0
d
v P u o e = = = = =
Operation poles;
(9)
1 2
3
-0.066 + 1.255i; -0.066 - 1.255i;
-0.156
p p
p
= =
=


PID Controller Desing
The PID controller is probably the most used feedback
control design. PID is an acronym for Proportional-Integral-
Derivative, referring to the three terms operating on the
error signal to produce a control signal. If u(t) is the control
signal sent to the system, y(t) is the measured output and
y
r
(t) is the desired output, and tracking error e(t) = y
r
(t) y(t),
a PID controller has the general form,
(10)

f
c d
i
f
d
1
( )
1
y = y
1 /
dy
u k e edt T
T dt
sT N
= +
+
}

PID controller parameters can be determined by closed-
loop Ziegler Nicholas tuning formula method [22, 23, 24].
Then, we calculate the PID controller parametres by using
the constants k
u
and t
u
with the Zigler - Nicholes closed-loop
oscillation method,
(11)
u u
5 and 12
7.2, 2.88, 43.2
t k
P I D
= =
= = =


Table 1. Ziegler-Nichols tuning Formula [24]
Controller PID PID
Proportional gain k
c
=0.6 k
u
k
c
=0.6*12
Integral time T
i
=0.5 t
u
T
i
=0.5*5
Derivative time T
d
=0.125 t
u
T
d
=0.125*5

Pole Placement
Block diagram of pole placement is given in Fig. 2. If the
linearized system considered is completely state
controllable, then poles of the closed-loop system may be
placed at any desired locations by means of state feedback
through an appropriate state feedback gain matrix K [25].
By using Ackerman method State feedback gain matrix
K can be obtained as follows (Kailath,1983):
(12)
1
1 1
1
2 1
+ +...............+ + I
=[0 0 0.....0 1]
=[ ...... ]
( )
n n
n n
c
n
T
n
o o o
o

=
=
T
n
-1
c
(A) A A A
q Q
Q B AB A B A B
K q A


PRZEGLD ELEKTROTECHNICZNY (Electrical Review), ISSN 0033-2097, R. 88 NR 7a/2012 35

Fig. 2 Block diagram of pole placement

Then from Fig. 2, for feedback linearized system we can
write,

(13) ( ) ( ) t t = u Kx

modifies the system equation to,

(14) ( ) ( ) ( ) ( ) t t r t = + x A- BK x B

the desired closed-loop poles are to be at,
(15)
1, 2 n
1 2
s ., ,
( )( )...........(s- )
n
s s


= .
= sI - A+BK


Online pole placement method is developed using
Taylor series expansion. Each point is linearized and then
feedback matrix K is caculated. However this takes a long
time and causes delay in the control. To speedup the
process, online pole placement using ANN is developed.

Online ANN Based Pole Placement
ANN based pole placement approach is created to
speed up the pole placement online data calculations. The
feedback gain matrix K can be calculated in a short period
of time by ANN. The ANN based pole placement controller
structure is obtained by using the matrix table K which is
obtained for the different values of Pd (0.48 - 1.0pu) in
online pole placement system. 20 training values and 5
testing values are used from the table.
To construct an adaptive controller such that the closed
loop poles remain at a fixed position, the system as well as
controller parameters are updated at each instant of time.


Fig. 3 Proposed Online ANN based Pole Placement

Fig. 3 represents the adaptive feedback Online ANN
based Pole Placement control architecture. The plant output
y(t) is a response of control input u(t) and disturbances P
d

(t).
The components of the input pattern consisted of the
control variables of the machining operation (power),
whereas the output pattern components represented the
measured factors (pole placement matrix K). The nodes in
the hidden layer were necessary to implement the nonlinear
mapping between the input and output patterns. In the
present work, 1-input, 10-hidden layers, 3 output layer
feedback propagation neural network has been used.
The Neural Network Toolbox under MATLAB used had
been here. The structure of the proposed neural Networks
used for feedforward back propagation (FBP) is shown in
Fig. 4. We trained the neural networks until the error
function was less than 1x10
-4
. Other neural networks
parameters are gradient 3.08x10
-6
, mu 1x10
-6
, learning rate
0.6, momentum 0.8 .


Fig. 4 The structure of FBP

In the present study, a hyperbolic tangent sigmoid and
pure linear functions given in below is employed as an
activation function in the training of the network.
(16)
( ) tanh ( )
( )
x x
x x
e e
f x x
e e
f x x

= =
+
=
and

where x in denotes weighted sum of the inputs. The
maximum number of epochs is arbitrarily kept at 1000.
Resulting of training regression and the Matlab/
SIMULINK ANN based online pole placement structure is
shown in Fig. 5 and Fig. 6, respectively.




















Fig. 5 Training regression





ANN based
K
Nonlinee
r System
Pole
Pament
Controller
r(t) y(t)
+
-
u(t)
P
d
(t)
36 PRZEGLD ELEKTROTECHNICZNY (Electrical Review), ISSN 0033-2097, R. 88 NR 7a/2012

Fig. 6 Matlab/SIMULINK model using online ANN based pole placement approach

Simulation Results
Simulation studies are carried out according to the
constant variation of the demand power in Table 2.

Table 2. Simulation parameter
Time
(second)
Demand power
P
d
(pu)
0 0.6
40 0.7
60 0.73
80 0.75
100 0.73
110 0.78
120 0.85
130 0.8

At each operating point, the generator terminal voltage
is maintained at 0.8465 pu.
K matrix components vary as seen in Fig. 7 in online
ANN based pole placement system.

0 200 400 600 800 1000 1200 1400 1600
-1
-0.5
0
Calculated of the K matrix (150 Sec.)
A
m
p
l
i
t
u
d
e

(
p
u
)


k1
k2
k3

Fig. 7 K-matrix components change with online
ANN based pole placement

The K matrix components are calculated them
constantly during the 150-second simulation. ANN of the
system are calculated with the help of the feedback matrix
K against the changing demand powers. 1501 variable K
components are calculated during the 150-second
simulation.
Control variable susceptance variation, variation of
output voltages, variation of PV curves, performance
analysis of temporary state and variation are given in Fig. 8,
Fig. 9, Fig. 10, Table 2 and Table 3, respectively.



















Fig. 8 SVC susceptance values











Fig. 9 Output voltage













Fig. 10 PV curves of no control, proposed algorithm and PID
control

The settle time with ANN based pole placement control
is very short while requested power (P
d
) is constant and the
settle time with PID took a long time as it can be seen in Fig
9. at the output voltages in 40 seconds.



0 50 100 150
-0.2
-0.1
0
0.1
0.2
0.3
time (s)
B
s
v
c

(
u
(
t
)
-
s
u
s
e
p
t
a
n
c
e
)

(
p
u
)


Online ANN pole placement
PID controller
0 50 100 150
0.5
0.6
0.7
0.8
0.9
1
1.1
time (s)
P
d

(
p
u
)
,








V
2

(
p
u
)


No control
Demand power (Pd)
Online ANN pole placement
PID controller
0.6 0.65 0.7 0.75 0.8 0.85
0.7
0.75
0.8
0.85
0.9
Pd (pu)
V
2

(
p
u
)


No control
Online ANN pole placement
PID controller
PRZEGLD ELEKTROTECHNICZNY (Electrical Review), ISSN 0033-2097, R. 88 NR 7a/2012 37
Table 3. The amount of overflow for the best conroller values
maximum overshoot, rise and setting time and steady state error
Controller Mp r
s
s
t
t=40s for
P
d
=0.7
pu, s
t

e
PID 0.0226 18.4 39.07 2.01 0.030
Online
ANN based
pole
placement
0.0171 4.6 7 4.8 0.005

Conclusion
Online ANN based pole placement method is proposed
in this study. The control of SVC by PID and proposed
method is studied for voltage stability enhancement. The
online pole placement controller developed for SVC is
designed with Taylor series expansion and Acermann
method. Output voltage and PV curve simulations show that
PID and proposed control algorithm delay the collapse. The
online ANN based pole placement shows the best
performance in point of the performance criteria (maximum
overshoot, best rise and setting time and steady state error)
shown in Table 3.

REFERENCES
[1] Anagquist, L., Lundin, B., and Samuelsson, J., Power
oscillation damping using conrolled reactive power
compensation, IEEE Trans, On Power System, (1993),
vol.8,no.2,pp 687-700, 1993
[2] Samir, B., Al, A., Design Of A Robust Svc Damping Controller
Using Nonlinear H Technique, The Arabian Journal For
Science And Engineering, (2005), Vol.30, No.1b
[3] Eslami, M., Shareef, H., Mohamed, A., Khujehzadeh, M.,
Particle swarm optimization for simultaneous tuning of static
var compensator and power system stabilizer, Przeglad
Elektrotechniczny (Electrical Review), (2011), 09a, pp 343-347
[4] Jordan, AJ., Linearization of non-linear state equation, Bulletin
of The Polish Academy of Sciences Technical Science, (2006),
Vol.54, No.1
[5] Wang, Y., Chen, H., Zhou, R., A nonlinear controller design for
SVC to improve power system voltage stability, Electrical
Power and Energy Systems 22(2000), 463470
[6] Conga, L., Wanga, Y., Hill, DJ., Transient stability and voltage
regulation enhancement via coordinated control of generator
excitation and SVC, Electrical Power and Energy Systems,
27(2005), 121130
[7] Astrm, K.J., and Hagglund, T., Revisiting the Ziegler-Nichols
step response method for PID control. Journal of Process
Control 14(2004), 635650
[8] Astrm, K.J., and Hagglund, T., Advanced PID Controlor, ISA-
Instrumentation, Systems, and Automation Society, (2006),
USA
[9] Chang, Y., Xu, Z., A novel SVC supplementary controllers
based on wide area signals, Electr Power Syst, Res 77(2007),
15691574
[10] Farrag, MEA., and Putrus, G.A., An on-line traning radial basis
function neural network for optimum operation of the UPFC,
European Transaction on Electrical Power , 21(2011), 27-39
[11] Arvanitis, K.G., Transactions of ttie ASME, (1999), 668 / Vol.
121
[12] Kumar, R., Khan, M., Journal of Vibration and Acoustics,
(2007), Vol. 129 / 601
[13] Shakir, M.S., Ahmad, M.N., and Amat, A., World Academy of
Science, Engineering and Technology 50 (2009)
[14] Hu, W.S., The Theory of Neural Network and Its Applications in
Engineering, (2006), Beijing: SinoMaps Press
[15] Yuan, C.R., Artificial Neural Network and Its Applications,
(1999), Beijing: Tsinghua University Press
[16]Wang, L.G., et al., Combined ANN prediction model for failure
depth of coal seam floors, Mining Science and Technology
19(2009), 06840688
[17] Said, .K., and Pirouti, M., Neural network-based load balancing
and reactive power control by Static VAR compensator,
International Journal of Computer and Electrical Engineering,
(2009), Vol.1, No.1, pp.25-31
[18] Al-Alawi, S.M., and Ellithy, K.A., Tuning of SVC damping
controllers over a wide range of bad models using an artificial
neural network. International Journal of Electrical Power &
Energy Systems, (2000), Vol. 22, Issue 6, pp.405-420
[19] Filik, U.B, Kurban, M., A new Approach for the short-Term Load
Forecasting with Autoregressive and Artificial neural Network
Models. International Journal of Computational Intelligence
Research, (2007), Vol. 3, No.1,pp-66-71
[20] Girish, JK., Artificial neural network application, I.A.R.I., (2007),
New Delhi,110012
[21] Kse, E., Abac, K., Aksoy, S., and Yaln, M.A., The
Comparison of the improving Effects of ULTC and SVC on
Dynamical Voltage Stability Using Neural Networks. IEEE
Modern Electric Power Systems MEPS10, (2010), Wroclaw,
Poland
[22] Ziegler, J.G., Nichols, N.B., and Rochester, N.Y., Optimum
Settings for Automatic Controllers. Transactions of ASME,
(1942), Vol.64, pp. 759-768
[23] Astrm, K.J., and Hagglund, T., Automatic Tuning of Simple
Regulators with Specifications on Phase and Amplitude
Margins. Automatica, (1984), Vol.20, No.5, pp. 645-651
[24] Hang, C.C., Astrm, K.J. and Ho, W.K., Refinements of the
Ziegler-Nichols tuning Formula, IEE Proceeding-D, (1991),
Vol.138, No.2
[25] Kailath, T., Linear Systems, (1983), Prentice-Hall Inc., Tokyo



Authors:
Lecturer. Ercan Kse, Department of Electronics and Computer
Education, Faculty of Technical Education, Mersin University,
33480,Tarsus, Mersin,Turkey, ekose@mersin.edu.tr; assist.prof.dr.
Kadir Abac, Department of Electronics and Computer Education,
Faculty of Technical Education, Mersin University, 33480,Tarsus,
Mersin,Turkey, kabaci@mersin.edu.tr; prof. dr. Saadettin Aksoy,
Department of Electrical and Electronics Engineering, Sakarya
University, Sakarya, Turkey, saksoy@sau.edu.tr.


The correspondence address is:

e-mail: ekose@mersin.edu.tr

You might also like