INTELLIGENT CONTROL OF FLEXIBLE AC TRANSMISSION SYSTEMS USING NEURAL
NETWORKS
Search of Engineering
Recommended by
Dr. Naeem M.S. Hannoon
Faculty of Engineering
Al-Madinah International University
Shah Alam – Malaysia
Dr. Naeem@mediu.edu.my
.
ABSTRACT: The paper presents the design of a new
intelligent control for flexible AC transmission systems
(FACTS) using a radial basis function neural network
(RBFNN). The architecture of this neural network consists of
a single neuron whose input is derived from the active or
reactive power or voltage derivation at the power system bus,
where the FACTS device ( in this case an unified power flow
controller) is located. The performance of this new single
neuron ANN controller is evaluated using both
single=machine infinite-bus and two-machine power system
subjected to various transient disturbances. The new
intelligent controller for FACTS exhibits a superior dynamic
performance in compensation to the existing PI control
schemes. Its simple architecture reduces the computational
overhead, thereby real-time implementation.
Kay words: INTELLIGENT CONTROL - FLEXIBLE AC
TRANSMISSION - SYSTEMS - USING - NEURAL
NETWORKS.
Introduction
With increased power transfer, transient and dynamic
stability are of increasing importance for secure operation of
power system. FACTS devices with a suitable strategy have
the potential to significantly improve the transient stability
margin. This allows increased utilization of existing network
closer to its thermal loading capacity, and thus avoiding the
need to construct new transmission lines. Amongst the
available FACTS devices, the UPFC (Unified Power Flow
Controller) is the most versatile one and can be used to
enhance system stability. The UPFC is capable of both
supplying and absorbing real and reactive power and consists
of two AC/DC converters. One of the two converters is
connected in series with the transmission line through a
series transformer and the other in parallel with the line
through a shunt transformer. The DC side of the two
converters is connected through a common capacitor which
provides DC voltage for the converter operation. Various
control strategies to control the series voltage magnitude and
angle and the shunt current magnitude have been presented in
the references [1-7]. The series converter voltage pharos can
be decomposed into in-phase and quadrate components with
respect to the transmission line current. The in-phase and
quadrate-voltage component are more readily related to the
reactive and real power flows in the transmission system.
During short-circuit and transient conditions, the decrease in
real power can be arrested by controlling the quadrate
component of the series converter voltage and hence the
improvement in transient stability. The series voltage in
phase component is either controlled by the reactive power
flow deviation or voltage deviation at the injected bus where
the UPFC is located.
The co-ordination of the real and reactive power control
schemes of the UPFC through an analytical approach is
difficult due to their interactions through the DC link
between the converters and also due to the nonlinear nature
of the problem. The PI regulator used for the purpose have
inadequacies of providing robust control and transient
stability over a wide range of power system operating
conditions
Use of ANNs (Artificial Neural Networks) for plant
identification and control is gaining interest [8-9]. In power
systems, the use of these controllers is increasing whereas
any worthy publication is not available for FACTS devices.
A potential advantage of the ANN is its ability to handle the
nonlinear mapping of the input-output space. The output of
the proposed controller is a neuron output which maybe
either the quadrate or the real voltage component of the series
inverter. The single neuron output will be a function of the
change either real power or change in the bus-voltage or
reactive power. This provides a nonlinear FACTS controller,
which can significantly improve the transient performance of
the power system. A similar neuron controller is used for the
shunt current if both series and shunt converter are controller.
The developed neuron controller (radial basis function
neuron network) for the FACTS is for a single-machine
infinite-bus and two- power system subjected to a wide range
of transient disturbances.
System Model
To study the new control strategy for the UPFC, both singlemachine infinite bus and two-machine system are considered
for transient simulations. The power systems are shown in
Bus1
Bus2
UPFC.
G1
~
G2
~
~
Gen.
L1
F
L2

~ Esh
Fig.1. Single-machine
infinite-bus power system
Fig.2. Two machine power
system
Fig.1. & Fig.2.
.
rd
The synchronous generator is represented by a 3 order
machine model and the generator excitation system has a
simple automatic voltage regulator (AVR). The series
converter injected a variable voltage source VC and the shunt
converter is a variable current is.
1.
Neural Network based Controller for UPFC
It is well known that a multi-layer feed forward neural
network with back propagation (BP) training algorithm is the
most widely used ANN model for nonlinear control of a
power system. However, the PB algorithm does not have a
mechanism to detect when the operating state of the power
system falls in a region with no training data. This produces s
serious drawback of the BP algorithm, as the operating
condition covers a wide range of system and fault conditions.
Recently researchers have begun to examine the use of radial
basis function (RBF) network for nonlinear control of plants
and systems as they offer a simple topological structure and
give insight as to how learning proceeds in an explicit
manner. Further for real-time implementation of a FACTS
controller, a single neuron RBF neuron network (RBFNN)
will be adequate.
Fig.3. shows the structure of the RBFNN, where the hidden
layer comprises a single neuron refereed to as the computing
unit
Hidden layer
x
Output layer
x
F(x)
ΔP(k)is Δ P(k)  ΔP(k)-ΔP(k-1 )
(2)
The quadrate voltage component is used in equation (1) as it
influences the real power error.
In a similar way, the weight adjustment for the single neuron
RBF controller to obtain the signal Vsp is done using the error
surface p(k) as
.
 q  C1q Q(k )  C2q  Q(k )  C3qVsp (k  1)
(3)
and Q(k)  Qref  Q1(k )
.
 Q(k )  Q(k )  Q(k  1)
(4)
Instead of reactive power deviation, the voltage deviation at
bus-1 can be taken as a control input. Since an error surface
is used to update the parameters of the controller any network
training is not required for generating the control voltage
components Vsp and Vsq. Further the constants C1p, C2p, C3p,
etc. are to be chosen suitably for the best performance of
UPFC in stabilizing power system oscillations.
In case the PI regulator is used in controlling Vsp and Vsq, the
equations are:
Qref
+
Q
-
(x)
Z-1
0
-
C2q
+
+
+ +
Fig.3. Single-neuron RBFNN structure
C3q
C1
q
In order to utilize a limited knowledge of the power system
dynamics to design the RBFNN controller for the series
converter (control of voltage outputs Vsp and Vsq) a
reinforcement method of weight adaption in used in this
paper. Fig.4, shows the schematic diagram of the proposed
controller for Vsp and Vsq respectively (Z-1 in the shift
operator). Vsp and Vsq are expressed as

V sp  Vcd sin   V q cos  Vcq sin  , and   tan -1 i cd i cq

Z-1
RBFN
N
Pref
+
-
Z-1
In obtaining the control signal Vsq, the input to the signal
neuron RBF controller is the
real power deviation signal P and weight adjustment is done
in using the error surface p(k) as
 p (k )  C1p P(k )  C2 p P(k )  C3 pVsq (k 1)
plant
P
Vcd , Vcq and icd , icq are the d-q axes component of the
injected series voltage and the line current ic.
Vsp
-
C2p
+
+ +
+
C3p
C1
p
Z-1
(1)
in the above equation
P(k )  Pref  P1 (k ), P1 (k )
is the real power flowing into transmission line at the kth
instant. The derivative term is
RBFN
N
Vsq
Fig. 4. RBFNN Controller
plant
Kiq 

Q,
Vsp   K pq 
P 

2.
Kip 

P
Vsq   K pp 
P 

(5)
Fig. 8 shows the transient response for the 3-phase fault case
from which it can be seen that the performance is found to be
excellent in the case of control of both the converters of
UPFC. The first swing excursions and settling time are
improved using the control on both the converters.
Computational Results
The single-machine infinite-bus power system shown in
Fig.1. and two machine power system in Fig.2. are taken for
digital simulation studies. The UPFC control scheme consists
of controlling voltage components Vsp and Vsq by using real
and reactive power deviations or real power or voltage
deviations. The current of the shunt converter is obtained
from the power balance equations at every control instant.
The following large disturbance cases are considered for
evaluating the performance of these controllers:
A. Fault Studies On Single-Machine Infinite-Bus Power
system
Case 1: P-Q control (only series)
C1p=.15; C2p=.15; C3p= -.05; C1q=.01 ; C2q=.01 ; C3q= -1,
Kpp=0.3 ; Kip=3; Kpp=.1; Kiq=1
The synchronous generator is assumed to operate at low
power output condition (P=0.1 p.u, and Q=0.2 p.u.) and a 3
phase fault occurs near the infinite-bus. The fault is cleared
after 0.1 sec. Of its inception and Fig. 5 shows the transient
response of the system with a conventional PI regulator for
the UPFC and the new single neuron RBF controller. From
the results it can be seen that the significant first swing
stability is achieved by using the new neural network
controller in comparison to the conventional PI controller.
The d.c. voltage excursions are also rapidly stabilized using
the new controller and this is a very important criterion for
successful operation of series and shunt voltage inverters.
The operating level of the generator is then changed to a high
power case with P = 1.4 p.u., and Q = 0.6 p.u. and a 3-phase
fault is initiated near the infinite-bus and cleared after 0.1
second of its inception. Fig. 6 shows transient response of
the power system for this operating condition with either PI
or the RBF neural controller. From the response, it can be
ascertained that the electromechanical are damped very
quickly in case of the new controller proving its superiority
over the conventional PI controllers used for the UPFC
control.
Case 3 : Both series and shunt control
C1p=.21; C2p=.21; C3p= -.03; C1i=.01; C2i=.01, C3i= -1
In this case the effectiveness of both series and shunt
converter controls of the UPFC are tested to damp
electromechanical oscillations of a single-machine infinitebus power system subjected to 3-phase fault. The power
loading of the generator is kept at P = 1.4 p.u. and Q = 0.6
p.u. and the real power deviation is used to control Vsq (Vsp is
not controlled) for the series converter. The shunt converter
on the other hand is used to control the shunt current I Q from
the UPFC-bus voltage deviation. The error surface used
for this control is
.
 i  C1i V (k )  C 2i  V (k )  C 3i iQ (k  1).
B. Two-Machine Power System
The 2-machine power system shown in Fig. 2 is used to study
the interaction between the generators with the UPFC located
on one of the lines of a parallel transmission system. The
real and reactive power injections at bus 1 are 0.9082 p.u.
and 0.4186 p.u. and at bus 2 are 0.3634 p.u. and 0.3435 p.u.
respectively. A 3-phase fault is initiated at the point F and
cleared in 0.1 sec.
For the UPFC control, a single-neuron RBF controller is used
as in the case of single-machine infinite-bus. A modulating
signal based on either the speed deviation between the two
generators or the UPFC bus angle  is used to improve the
damping. In that case the signal P (change in real power in
the transmission line after UPFC from the reference) is to be
replaced by P + K. (1 - 2)
Fig. 9 shows the performance of the 2-machine power system
in which the RBFNN controller is found to provide
significant improvement in damping inter area oscillations in
comparison to the conventional PI regulators. The first
swing overshoot is significantly reduced in case of the new
UPFC controller.
5.
Conclusions
This paper presents a new single neuron Radial Basis
Function Controller (RBFNN) for the UPFC control in a
single machine-infinite-bus and two-machine power system.
The single neuron controller uses either the real and reactive
power deviations or the real power and voltage deviations at
the UPFC injection bus to provide the magnitude and phase
angle of the variable voltage source for the series converter.
The shunt converter current can be similarly controlled using
a RBFNN and the UPFC with both series and shunt controls
provides the best transient stability performance of the power
system. The parameters of the single neuron RBF controller
are adjusted using an extended Kalman filter. More than one
neuron has also been used for providing control voltages,
however, the improvement in transient stability performance
has been minimal in comparison to the single neuron case.
The simple architecture of the controller has the potentiality
of implementation on a DSP chip in real-time environment.
REFERENCES
1. L. Gyugyi, C.D. Schauder, S.L. Torgerson and A. Edris,
“The Unified Power Flow Controller: A new approach to
Power Transmission Control”, IEEE Trans. On Power
Delivery, Vol. 10, No. 2, 1995, pp. 1088-1097.
2. L. Gyugyi, “Unified Power flow concept for flexible AC
Transmission systems”,
IEE Proc.-C, Col. 139, No.4,
1992, pp. 323-332.
3. C.D. Schauder and H. Mehta, “Vector Analysis and
Control of Advanced Static VAR Compensator”, IEE Proc.C, Vol. 140, No. 4, 1993, pp. 299-306.
4. M. Noorzian, L. Angquist, M. Ghandari and G.
Anderson, “Improving Power System Dynamics by SeriesConnected FACTS Devices”, IEEE Trans. on Power
Delivery, Vol. 12, No. 4, 1997, pp. 1635-1641.
5. M. Noorzian and G. Anderson, “Damping of Power
System by Controllable components”, IEEE Transactions on
Power Delivery, Vol. 9, No. 4, 1994, pp. 2046-2054.
6. K. R. Padiyar and A. M. Kulkarni, “Control design and
simulation of Unified Power flow controller”, IEEE Trans.
on Power Delivery, Paper PE-172-PWRD-0-12-1997.
7. S. Limyingcharoen, U. D. Annakkage and N. C.
Pahalawaththa, “Fuzzy logic based Unified Power flow
controllers for transient stability improvement”, IEEE Proc.C, Vol. 145, No. 3, 1998, pp. 225-232.
8. K. S. Narendra, K. Parthasarathy, “Identification and
control of dynamical systems using Neural Networks”,
IEEE Trans. on Neural Networks. Vol. 1, No. 1, March 1990,
pp. 4-27.
9. K. G. Narendra, K. Khorasani, V. K. Sood, R. V. Patel,
“Intelligent Current Controller for HVDC Transmission
Link”, IEEE Trans. on power Systems, Vol. 13, No. 3, Aug.
1998, pp. 1076-1083.