American Journal of Electrical Power and Energy Systems

2014; 3(6): 101-106

Published online November 10, 2014 (http://www.sciencepublishinggroup.com/j/epes)
doi: 10.11648/j.epes.20140306.11
ISSN: 2326-912X (Print); ISSN: 2326-9200 (Online)

Research methods for power system stability using

Adaptive Neural Fuzzy Inference Systems
Phan Xuan Le1, 2, *, Nguyen Le Thai1, 2, Nguyen Le Minh Tri1, 3
1
Faculty of Mechanical and Electrical Engineering, Kunming University of Science and technology, Kunming City, Yunnan Province, China
2
Faculty of Electric and Electronic Engineering, Tuy Hoa Industrial College, Tuy Hoa City, Phu Yen Province, Vietnam
3
Faculty of Electrical Engineering, Kien Giang Technology and Economics College, Kien Giang Province, Vietnam

Email address:
phanxuanle.ts@gmailcom (P. X. Le), thai.nguyenle.tt@gmailcom (N. L. Thai), [email protected] (N. L. M. Tri)

To cite this article:

Phan Xuan Le, Nguyen Le Thai, Nguyen Le Minh Tri. Research Methods for Power System Stability Using Adaptive Neural Fuzzy Inference
Systems. American Journal of Electrical Power and Energy Systems. Vol. 3, No. 6, 2014, pp. 101-106. doi: 10.11648/j.epes.20140306.11

Abstract: The performance of the Automatic Voltage Regulate (AVR) and the Power System Stability (PSS) methods may be
degraded stability of the power system. This paper presents an Adaptive Neural Fuzzy Inference Systems (ANFIS) algorithm for
stability of the power system, we use an Adaptive Network based Fuzzy Interference System architecture extended to response
with multivariable systems. By using a hybrid learning method, the suggested ANFIS can setting structure diagram input - output
based on both human knowledge and stipulated input-output data pairs. Simulation results present the convergence of the
algorithm is improved.

Keywords: AVR, PSS, ANFIS

ANFIS algorithm.
1. Introduction This paper refers to a controller using Adaptive Neural
Fuzzy Inference Systems to replace the power system
In power systems, the electromechanical oscillations of the stabilizer (PSS). This is one control algorithm is used widely
generators may adversely affect. Thus, using the power in the field of automatic control which is effective for the
system stabilizer PSS controllers is necessary. It will damp nonlinear of plants. Adaptive Neural Fuzzy Inference
the electromechanical oscillations in power systems. The Systems algorithm is used for power system operation which
plants of linear models at each operating point are different. excitation system of the generator will be automatically
The proposed PSS based primarily on a transfer function and adjusted to limit disturbance. In there, research issues set for
a linear model of the plant has been widely used [1, 2, 3]. controller parameters is very important, it determines the
However, the feature of the power systems has dynamic performance characteristics of the system generator.
and highly nonlinear. Therefore, the performance of PSS can The remainder of the paper is organized as follows.
degrade under variations of the nonlinear characteristics of Section II describes Mathematical model of the system to
the plant. synchronize generators. In section III, The Adaptive Neural
In recent years researchers used neural - fuzzy techniques Fuzzy Inference Systems (ANFIS) in Small-signal Stability
to control complex systems utilizing solely the input-output for Power System is presented. Section IV presents
data sets. Meanwhile, fuzzy control technique requires simulation results. The conclusions are given in section V.
human knowledge and experience to set the IF – THEN rules.
This deficiency can be overcome by combining the neural
networks and fuzzy logic, the proposed ANFIS with 2. Mathematical Model of the System to
rule-based of controllers has shown promising results [ 5, 6]. Synchronize Generators
The viewpoint proposed here which is Adaptive Neural
Fuzzy Inference Systems is used to solve the problems For small-signal stability analysis, dynamic modeling is
mentioned above. We use ANFIS replace the conventional required for the major components of the power system. It
control method to solve the performance concerns. In this includes the synchronous generator, excitation system,
study, a first step is taken towards systematic analysis using automatic voltage regulator (AVR), etc. The model shown in
102 Phan Xuan Le et al.: Research Methods for Power System Stability Using Adaptive Neural Fuzzy Inference Systems

Figure 1 is used to obtain the linearized dynamic model [13].

(2)

Therefore, the model synthesis of linear systems has been

derived from power system disturbances as Figure 2.

Figure 1. Single machine connected to a large system through transmission

lines

The governing dynamic equations of the system [9] are

shown as in (1).

 ∆ Te = K 1 ∆ δ + K 2 ∆ ψ fd

∆ψ = K 3 ∆E − K ∆δ 
 1 + pT3  
fd fd 4


 ∆ E t = K 5 ∆ δ + K 6 ∆ ψ fd (1)

 p ∆ ω r = 1 ( ∆ Tm − K S ∆ δ − K D ∆ ω r )
 2H
 p∆δ = ω ∆ω
 0 r

Where
KS is synchronizing torque coefficient in pu torque/rad
KD is damping torque coefficient in pu torque/pu speed
deviation
H is inertia constant in (MW-Sec/MVA)
∆ωr is speed deviation in pu = (ωr − ω0 ) / ω0
∆δ is rotor angle deviation in elec. rad
ω0 is rated speed in elec. rad/s = 2π f 0
The coefficients K1 ->K6 depend on the parameters of the
grid and the power system voltage. Figure 2. Block diagram representation with Exciter and AVR.

∆Te 
K1 =
∆δ ψ fd =ψ fd 0 3. The Adaptive Neural Fuzzy Inference
= n1 (ψ ad 0 + Laqsid 0 ) − m1 (ψ aq 0 + L i ) '
ads q 0
Systems (ANFIS) In Small-Signal
Stability for Power System
∆Te 
K2 =  The Adaptive Neural Fuzzy Inference Systems is a kind of
∆ψ fd 
δ =δ 0 neural network which is based on Takagi–Sugeno fuzzy
L'ads inference system. Since it integrates both neural networks and
= n2 (ψ ad 0 + Laqs id 0 ) − m2 (ψ aq 0 + L'adsiq 0 ) + i q0 fuzzy logic principles, it has ability to capture the advantages
L fd
of both in a single framework. Its inference system
corresponds to a set of fuzzy IF–THEN rules that it is capable
L fd  1 
K3 =   of learning to approximate nonlinear functions [10]. Hence,
' '
Ladu 1 − ( Lads / L fd ) + m2 Lads  ANFIS is regarded to be a universal estimator.[8]
In this section, we propose a class of adaptive networks
m1L'ads which are functionally equivalent to fuzzy inference systems.
K4 =
L fd Ladu The proposed architecture this is called ANFIS, standing for
Adaptive Neural Fuzzy Inference Systems. We describe how
ud 0 u to analyse the parameter set in order to apply the hybrid
K5 = [ − Ra m1 + X d n1 ] + q 0  − Ra n1 + X d' m1  learning rule. Besides, we demonstrate how to apply the
Ut0 Ut0
u 
Stone-Weierstrass theorem to ANFIS will be simplified [4, 5]
ud 0 L' 
K6 = [ − Ra m2 + X d n1 ] + q 0  − Ra n2 − X d' m2 + ads  Fuzzy if-then rules and how the radial basis function
Ut 0 U t 0  L fd 
network relate to this kind of simplified ANFIS
For simplicity, we assume the fuzzy inference system under
On the other hand, excitation system is used ST1A which is
consideration has two inputs x = ∆ω and y = ∆Pe and one
synthesized in the form of small disturbance in the literature [7,
output f. Suppose that the rule base includes two fuzzy if-then
9, 11]:
 American Journal of Electrical Power and Energy Systems 2014; 3(6): 101-106 103

rules of Takagi and Sugeno’s type [6, 10].

In this study using 5 set of fuzzy rule, such as:

For the training of the network, there is a forward pass and a

backward pass. We now look at each layer in turn for the
forward pass. The forward pass propagates the input vector
through the network layer by layer. In the backward pass, the
error is sent back through the network in a similar manner to Figure 4. ANFIS architecture
back-propagation.
Then the fuzzy reasoning is illustrated in Figure 3. And the In this study we choose (3) to be bell-shaped with maximum
corresponding equivalent ANFIS architecture is in as Figure4. equal to 1 and minimum equal to 0, such as (4) and (5).

(4)

(5)

Where {aij, bij, cij} is the parameter set. As the values of

these parameters change, the Bell-shaped functions vary
accordingly, thus exhibiting various forms of membership
functions on linguistic label Aj, Bj. Parameters in this layer are
referred to as premise parameters.
- Layer 2: Every node in this layer is a circle node labeled II
which multiplies the incoming signals and sends the product
out, as in (6).

(6)

Figure 3. Fuzzy reasoning

The node functions in the same layer are of the same

function family as described below:
- Layer 1: Every node in this layer is a square node with a
node function.

(3)

Where: i = 1 ÷ 2 , j = 1 ÷ 5 , x is the input to node i and Aj is

the linguistic label (small , large, etc.); associated with this
node function. In other words Oi j is the membership function Each node output represents the firing strength of a rule, (In
of Aj , and it specifies the degree to which the given x satisfies fact, other T-norm operators that perform generalized AND
can be used as the node function in this layer).
the quantifier Aj .
- Layer 3: Every node in this layer is a circle node labeled N.
The i-th node calculates the ratio of the i-th rule’s firing
strength to the sum of all rules’ firing strengths, as in (7):

(7)
104 Phan Xuan Le et al.: Research Methods for Power System Stability Using Adaptive Neural Fuzzy Inference Systems

For convenience, outputs of this layer will be called Table 1. Synchronous Machine Parameters
normalized firing strengths Symbol Parameters Value
- Layer 4: Every node i in this layer is a square node with a R1 Stator winding resistance at 150C 0.0077 pu
node function: R2 Rotor winding resistance at 150C 0.126 pu
xd Direct axis synchronous reactance 1.0494 pu
(8) xq Quadrature axis synchronous reactance 0.648 pu
x’d Direct axis transient reactance 0.2887 pu
Where w j is the output of layer 3, and {pj, qj, rj} is the x’’d Direct axis over transient reactance 0.191 pu
xe Stator leakage reactance 0.1244 pu
parameter set. Parameters in this layer will be referred to as
xq Quadrature axis synchronous reactance 0.648 pu
consequent parameters.
x″q Quadrature axis over transient reactance 0.197 pu
Excitation winding time constant when stator
T’do 6.88 s
winding open circuit
Cosϕ Rated power factor 0.85
H The coefficient of inertia 1.5
(9) S Rated capacity 1.0 pu
P Rated power 0.85 pu
Q Rated power 0.85 pu
Ut Rated voltage 1.0 pu
It Rated current 1.0 pu
- Layer 5 The single node in this layer is a circle node ω Rated angular speed of rotor 1.0
labeled E that computes the overall output as the summation of f Rated frequency 50 Hz
all incoming signals, i.e, as in (10):
(pu = Per unit, Hz = Hertz, s = second.)

(10) - With PSS-2A and excitation system is shown in Figure 5.

Where x, y is the input: x = ∆ω and y = ∆Pe ; Output

signal is in layer 5 f is the output, shown in Figure 4.
Hybrid Learning Algorithm
From the proposed ANFIS architecture in Figure 4, it is
observed that given the values of premise parameters, the
overall output can be expressed as a linear combinations of the
consequent parameters. More precisely, the output f in Figure 4
can be rewritten as in (11).
Figure 5. PSS-2A and excitation system

The IEEE Type-ST1A dynamic equation [9] with PSS-2A

is shown as in (12):

1
Eɺ fd =  K A (Vref + VPSS − Vt ) − E fd  (12)
TA
(11)
- With ANFFIS and excitation system is shown in Figure 6.
The IEEE Type-ST1A dynamic equation with ANFIS is
shown as in (13):

4. Simulation Results
In this section the ANFIS system is simulated using
Matlab-Simulink [12].
In power system analysis, the parameters are usually
convenient to user a per unit system to normalize system
variables. The simulation parameters are selected as following Figure 6. ANFIS and excitation system
Table 1 in literature [14].
 American Journal of Electrical Power and Energy Systems 2014; 3(6): 101-106 105

1
Eɺ fd =  K A (Vref + VANFIS − Vt ) − E fd  (13)
TA

The simulation results:

At the time t=0s, the generator is connected with the power
system. We compare the performances of three different
approaches: 1) the power systems using traditional AVR. 2)
The power systems using PSS algorithm. 3) The power
systems using ANFIS algorithm. Figure 7 and Figure 8 show
simulation results of the characteristics of rotor speed. Figure
8 gives rotor speed using ANFIS algorithm has fluctuated in
2.2 seconds (from 0s to 2.2s) and stable with ω=1pu.

Figure 9. The characteristic of angular speed of rotor, the load power of

300MW is connected into the power system after time 8s

Figure 10, Figure 11 and Figure 12 give the simulation

results of the characteristic of available power under three
different methods.
Figure 11 shows power of the power system using ANFIS
algorithm has fluctuated in 2.3 seconds (from 0s to2.3s) and
stable with Pe=0.85pu at time t = 0s the generator is connected
to power system.

Figure 7. The characteristic of angular speed of rotor, the generator is

connected into the power system at time t=0s

Figure 10. The characteristic of angular speed of power, the generator is

connected into the power system at time t=0

Figure 8. The characteristic of angular speed of rotor zoom in Figure 7

Figure 8 also gives power systems using ANFIS algorithm

works better than AVR and PSS algorithm.
At time t = 8s, we add connected load 300MV to power
system. Figure 9 show that simulation results of the
characteristics of rotor speed under three different methods.
Figure 9 show that the rotor speed has fluctuated in 2.23
seconds (from 8s to 10.23s) and stable with ω=1pu. Figure 9
also show that the performance of the power system using Figure 11. The characteristic of angular speed of power, the generator is
ANFIS algorithm is superior to another two methods. connected into the power system zoom in Figure 10
106 Phan Xuan Le et al.: Research Methods for Power System Stability Using Adaptive Neural Fuzzy Inference Systems

[2] P.demello and Charles concordia, “Concept of Synchronous

Machine Stability as Affected by Excitation Control,” IEEE
transactions on power apparatus and systems, vol. pas-88,no. 4,
pp. 316–329, 1969.

[3] S.E.M. de Oliveira, “Effect of Excitation Systems and of Power

Systems Stabilizers on Synchronous Generator Damping and
Synchronizing Torque,” IEE Proceedings, Vol 136, Pt.C, No5,
pp. 264–270, September 1988.

[4] R. Sivakumar, C. Sahana, P. A. Savitha, “Design of ANFIS

based Estimation and Control for MIMO Systems,” ISSN:
2248-9622, Vol. 2, Issue 3, pp. 2803–2809, May-Jun 2012.

[5] Jyh-Shing Roger Jang, “ANFIS: Adaptive-Network-Based

Fuzzy Inference System,” IEEE transactions on systems, man,
Figure 12. The characteristic of angular speed of power, the load power of and cybernetics, vol. 23, no. 3, pp. 665–685, may/june 1993.
300MW is connected into the power system after time 8s
[6] T. Takagi and M. Sugeno, “Fuzzy identification of systems and
its applications to modeling and control,” IEEE transactions on
After 8 seconds, we connect a load of 300MW into the
systems, man, and cybernetics, Vol. SMC-15, No.1, pp.
power system. Figure 12 illustrates power of the power system 116–132, January/February, 1985.
using ANFIS algorithm has fluctuated in 2.23 seconds (from
8s to 10.23s) and stable with Pe=0.85pu. [7] Carlos Ernesto Ugalde Loo, Luigi Vanfretti, Eduardo
Liceaga-Castro, Enrique Acha, “Synchronous Generators
Through the results in Figure 10, Figure 11, Figure 12, we
Modeling and Control Using the Framework of Individual
see that the power system stability using ANFIS algorithm can Channel Analysis and Design Part 1,” International Journal of
achieve better results than AVR and PSS methods. Emerging Electric Power Systems, Volume 8, Iss 5, Art 4, pp.
1–26, 2007.
5. Conclusion [8] A. Abraham, "Adaptation of Fuzzy Inference System Using
Neural Learning,” Springer-Verlag Berlin Heidelberg, 2005.
In this paper, an ANFIS algorithm has been presented for
stability control of the power system. The main contributions [9] P.Kundur, “Power System Stability and control,”
Vice-President, Power Engineering. Powertech Labs Inc.,
of the paper are that we use ANFIS algorithm for the problem
Surrey, British Columbia, 1993.
of small signal stability in power system which enhancing
damping of system oscillations via generator excitation control. [10] J-s-r.Jang, c-t.sun, e.mizutani, “Neuro-fuzzy anh soft
This study provides an alternative algorithm for power system computing,” Prentice Hall Upper Saddle river, NJ 07458, 1997
stabilizer PSS to reduce the response time of the rotor speed, [11] IEEE Recommended Practice for Excitation System Models
the response time of output power of the generator and for Power System Stability Studies, IEEE Std 421.5TM-2005.
reinforcing the power stability in the power system. Simulation
[12] Chee-mun Ong, “Dynamic Simulation of Electric Machinery
results show that the ANFIS algorithm can achieve better
using Matlab Simulink,” Prentice Hall PTR, 1998.
results than PSS and AVR methods.
[13] Sauer Peter W. and Pai M. A, “Power System Dynamics and
Stability,“ Pretice Hall, 1998.
References [14] Sichuan Dongfeng Electric Machinery Works Co., Ltd,
“Hydrogenerator Product Instructions, Product type:
[1] Balwinder Singh Surjan and Ruchira Garg, “Power System SF32.3-16/4500,” 2008.
Stabilizer Controller Design for SMIB Stability Study,” ISSN:
2249 – 8958, Volume-2, Issue-1, pp. 209–214, 2012.

View publication stats