WSEAS TRANSACTIONS on POWER SYSTEMS P. Pao-La-Or, T. Kulworawanichpong, And A.

Oonsivilai

Bi-objective Intelligent Optimization for Frequency Domain

Parameter Identification of a Synchronous Generator
P. PAO-LA-OR*, T. KULWORAWANICHPONG, and A. OONSIVILAI
Power and Energy Research Unit, School of Electrical Engineering
Institute of Engineering, Suranaree University of Technology
111 University Avenue, Muang District, Nakhon Ratchasima, 30000
THAILAND
*
corresponding author: [email protected]

Abstract: - This paper describes an intelligent approach to estimate parameters of a direct-axis equivalent
circuit of a synchronous generator from frequency response data using bi-objective intelligent optimization
methods, genetic algorithms and adaptive tabu search. The genetic algorithms and adaptive tabu search are
capable of finding a global minimum within a given search interval. The sum square error of magnitude and
phase of the d-axis equivalent circuit transfer function to formulate a bi-objective optimization problem is
minimized to best fit the measured data extracted from the frequency response test of the machine. As a result,
exploitation of the bi-objective optimization based on genetic algorithms and adaptive tabu search give very
good results than those of using either the magnitude or the phase as a single objective. This confirms the
effectiveness of the intelligent approach for solving bi-objective optimization problems described in this paper.

Key-Words: - Parameter Identification, Genetic Algorithms, Adaptive Tabu Search, Bi-objective Optimization,
Synchronous Generator, Frequency Response

1 Introduction aspiration criteria are keys. With some

To date under deregulated power market enhancement, back-tracking and adaptive search
environment electric utility has become increasingly radius mechanisms are included to speed up the
much more complex that the past. Apart from process and to guarantee the convergence. There
economic view point, stability problems are equally have been various researches and applications of GA
important to operate electric power system in real and ATS covering in most fields of studies.
time. To handle any stability-related problems Therefore, it would be good for solving this problem
accurate parameter estimation of a synchronous based on the GA and ATS.
generator is concerned in both direct and quadrature This paper illustrates the way to apply the genetic
models. Several kinds of tests are used to determine algorithms and the adaptive tabu search to solve a
the direct-axis equivalent circuit parameters. These bi-objective optimization problem in order to
include on-line tests [1], standstill frequency estimate a d-axis transfer function of a synchronous
response (SSFR) [2,3,4] and time domain [5] generator, which is explained in detail in section 2.
testing. From literature, the frequency response test Section 3 gives a brief of the step-by-step intelligent
has become one of the most popular approaches to parameter estimation based on the genetic
obtain the synchronous transfer function parameters. algorithms and adaptive tabu search. Section 4
With this method, the problem is reduced to find shows test results and discussion. The last section is
location of suitable poles and zeros of the machine the conclusion.
transfer function. To complete this task, an efficient
intelligent search method can be employed. Two
efficient intelligent search techniques, Genetic 2 Direct-Axis Model Structure of a
Algorithm (GA) and Adaptive Tabu Search (ATS), Synchronous Machine
are used to illustrate this identification. The GA is a The direct-axis of a synchronous machine includes
searching method based on two natural processes: two terminal ports. These correspond to the direct-
selections and genetics. It is considered as an axis equivalent armature winding and the field
evolutionary computation which has been proved to winding. The complete direct-axis equivalent circuit
be a very powerful optimization method in an which second order model referred to the stator is
artificial intelligence area of interest. The ATS is shown in Fig.1 [6]
also a stochastic search technique based on the local
search or hill-climbing principles. Tabu List and

Issue 3, Volume 3, March 2008 ISSN: 1790-5060

56
 WSEAS TRANSACTIONS on POWER SYSTEMS P. Pao-La-Or, T. Kulworawanichpong, And A. Oonsivilai

3 Bi-objective Intelligent Parameter

Estimation Based on Genetic
Algorithms and Adaptive Tabu Search
There exist many different approaches to identify
synchronous generator’s parameters. For the GA and
ATS are not new anymore. There exist a hundred of
works employing for these two intelligent searching.
Fig. 1. Direct-axis equivalent The GA is a stochastic search technique that leads
a population of solutions using the principles of
…, where genetic evolution and natural selection, called
Ll = armature leakage inductance genetic operators e.g. crossover, mutation, etc. With
Lad = stator to rotor mutual inductance successive updating new generation, a set of updated
solution gradually converges to the real solution.
L f 1d = mutual inductance between field winding and Because the genetic algorithms is very popular and
damper winding widely used in most research areas where an
L1d = damper winding leakage inductance intelligent search technique is applied, it can be
summarized briefly as follows [10].
R1d = damper winding resistance
L fd = field winding leakage inductance 1. Initialization: Randomly initialize populations or
chromosomes and set them as a search space and
R fd = field winding resistance
then evaluate their corresponding fitness value via
the objective function.
The direct-axis specifically operational 2. Evolution: Apply the genetic operators to create
inductance (OI) transfer function Ld(s) of an offspring population as the sequence below,
synchronous machine has the form given below a. Selection: Form a set of mating pool with the
[7,8]. It is the Laplace transform of the ratio of the same number of the population size by using a
direct-axis armature flux linkages to the direct-axis random procedure, e.g. the roulette-wheel or
current, with the field winding short-circuited. tournament schemes, with the assumption that each
chromosome has a different chance. The higher the
(1 + T1 s)(1 + T2 s ) fitness value, the higher the chance or probability.
Ld ( s ) = L d (1) b. Crossover: This operation is applied to a subset
(1 + T3 s)(1 + T4 s )
of the mating pool by taking a pair of chromosomes
called the parents. The parents will yield a pair of
Ld(s) is often expressed in terms of transient and offspring chromosomes. This operation involves
subtransient quantities used [9], exchanging sub-string of the parent chromosomes. It
is performed by choosing a random position in the
⎡ ⎛ L′d ⎞ ⎤ ⎡ ⎛ L′′ ⎞ ⎤ string and then swapping either the left or right sub-
⎢1 + ⎜⎜ τ ′do ′ d ⎟⎟ s ⎥
⎟⎟ s ⎥ ⎢1 + ⎜⎜ τ ′do
⎝ Ld ⎠ ⎦⎣ ⎝ L′d ⎠ ⎦ strings of this position (one-point crossover) with its
Ld ( s ) = L d ⎣ (2) chromosome mate.
(1 + τ′do s )(1 + τ ′do′ s ) c. Mutation: For the chromosome to be mutated,
the values of a few positions in the string are
…, where randomly modified. To prevent complete loss of the
L d = synchronous inductance (p.u.) genetic information carried through the selection
and crossover processes, mutation (if use at all) is
L′d = transient inductance (p.u.) limited to typically 2.5% of the population.
L′d′ = subtransient inductance (p.u.) 3. Fitness Test: Evaluate the fitness value for the
generated offspring population.
τ ′do = transient open-circuit time constant (secs)
4. Convergence Check: Check for violation of all
τ ′do
′ = subtransient open-circuit time constant (secs) termination criteria. If not satisfied, repeat the
evolution process.

The ATS is also a stochastic search technique

based on the local search or hill-climbing principles.
Tabu List and aspiration criteria are keys. With

Issue 3, Volume 3, March 2008 ISSN: 1790-5060

57
 WSEAS TRANSACTIONS on POWER SYSTEMS P. Pao-La-Or, T. Kulworawanichpong, And A. Oonsivilai

some enhancement, back-tracking and adaptive

search radius mechanisms are included to speed up
the process and to guarantee the convergence. The
adaptive tabu search can be briefly summarized, step
by step, as follows [11,12].

1. Initial Parameters: Initialize the Tabu List and

reset all program counters.
2. Initial Guess Selection: Randomly select an initial
solution from the search space and assign it as the
global minimum and also a recent updated solution. Fig. 2. Intelligent parameter identification
3. Neighborhood Creation: Create a searching sub-
2
space around the recent updated solution. Evaluate ⎛y
N
− y simulated ⎞
the objective function of all neighborhood members. SSE = ∑ ⎜⎜ measured ⎟
⎟
(3)
A solution that gives the minimum objective i =1 ⎝ y measured,max ⎠
function among them is given as the next-round
recent updated solution. with the bi-objective function
4. Tabu List Formulation: If the recent updated
solution is better than a global solution found thus f bi -obj = (Wmag ) ⋅ SSE mag + (Wphase ) ⋅ SSE phase (4)
far, keep the recent updated solution in the Tabu List
and set a new global solution as the recent updated
and
solution. Otherwise put the recent updated solution
in the Tabu List instead to prevent cycling. Update
(Wmag ) + (Wphase ) = 1.0 (5)
the global minimum.
5. Convergence and Aspiration Criterion Check:
Check the termination criteria and the aspiration …, where
criteria, respectively f bi -obj is the bi-objective function
a. Go to step 6 if the termination criteria is W is the weighted SSE of magnitude or phase
satisfied, otherwise repeat step 3.
on frequency response characteristics
b. Activate the adaptive search radius mechanism
if necessary to speed up the searching process. y measured is the measured magnitude or phase on
c. Activate the back-tracking mechanism if a local frequency response characteristics
minimum trap occurs and repeat step 3. y simulated is the simulated magnitude or phase on
6. Termination: Terminate the search process. The frequency response characteristics
last updated is the global minimum found.

In this paper, the GA and ATS are selected to

build up an algorithm to identify such parameters.
4 Results and Discussion
As shown in Fig.2, the whole process of this For intelligent identification, the GA and the ATS
intelligent parameter identification is summarized. are employed, respectively. The followings describe
Briefly, the procedure to perform the proposed parameter setting for these two intelligent search
identification is described as follows. First, used in this paper.
frequency responses of magnitude and phase based
on SSFR test data of a synchronous generator are GA:
measured. Second, the GA and ATS are employed Number of population = 50
to generate a set of initial random parameters. With Crossover probability = 70 %
the searching process, the parameters are adjusted to Mutation probability = 1.4 %
give response best fitting close to the test data. To
perform the searching properly, a bi-objective ATS:
function is the key. In this paper, the sum of squared Number of neighbourhood = 30
errors (SSE) [13] is used as shown in the following Initial search radius = 10 %
equation.

Issue 3, Volume 3, March 2008 ISSN: 1790-5060

58
 WSEAS TRANSACTIONS on POWER SYSTEMS P. Pao-La-Or, T. Kulworawanichpong, And A. Oonsivilai

Variable search spaces:

L d ∈ [0.50, 4.00]
L′d ∈ [0.05, 1.00]
L′d′ ∈ [0.01, 0.50]
τ ′do ∈ [1.00, 8.00]
τ ′do
′ ∈ [0.01, 0.10]

Termination criteria:
Maximum error allowance = 0.01 (SSE)
Maximum number of iteration = 1500

Table 1 also shows the parameters of turbine

generator, 555 MVA/24 kV/60 Hz/0.9 pf [14]
obtained by using the experimental and the two
a)
searching techniques with bi-objective by various
weighted SSE. The values appeared in this table is
the best of 30 trials.
Comparing with the experimental results and the
effectiveness and the accuracy of each method with
bi-objective by weighted SSE are revealed as shown
in Fig.3 - Fig.7.

Table 1. Comparison among obtained parameters

Parameters L d L′d L′d′ τ ′do τ ′do
′
Methods (p.u.) (p.u.) (p.u.) (secs) (secs)
Experimental 1.9700 0.2700 0.1270 4.3000 0.0310
GA,
mag1.00-phase0.00 2.1273 0.2845 0.1477 4.3111 0.0427
ATS,
mag1.00-phase0.00 2.4263 0.3201 0.1923 4.4259 0.0313
GA, b)
mag0.75-phase0.25 2.0991 0.2917 0.1428 4.4120 0.0355
ATS, Fig. 3. Frequency response characteristics of the
Mag0.75-phase0.25 1.7662 0.2575 0.1026 3.8661 0.0518 experiment and the searching techniques with bi-
GA, objective by weighted SSE (magnitude=1.00 and
mag0.50-phase0.50 1.8957 0.2530 0.1183 4.3265 0.0284 phase=0.00): a) GA, b) ATS
ATS,
mag0.50-phase0.50 2.7062 0.4747 0.2341 4.2679 0.0337
GA,
mag0.25-phase0.75 1.9803 0.2785 0.1295 4.4444 0.0315
ATS,
mag0.25-phase0.75 1.7921 0.2444 0.1062 4.2369 0.0431
GA,
mag0.00-phase0.10 2.2632 0.3082 0.1449 4.2615 0.0327
ATS,
mag0.00-phase0.10 5.9046 0.9175 0.4343 4.1843 0.0359

Issue 3, Volume 3, March 2008 ISSN: 1790-5060

59
 WSEAS TRANSACTIONS on POWER SYSTEMS P. Pao-La-Or, T. Kulworawanichpong, And A. Oonsivilai

a) a)

b) b)
Fig. 4. Frequency response characteristics of the Fig. 5. Frequency response characteristics of the
experiment and the searching techniques with bi- experiment and the searching techniques with bi-
objective by weighted SSE (magnitude=0.75 and objective by weighted SSE (magnitude=0.50 and
phase=0.25): a) GA, b) ATS phase=0.50): a) GA, b) ATS

Issue 3, Volume 3, March 2008 ISSN: 1790-5060

60
 WSEAS TRANSACTIONS on POWER SYSTEMS P. Pao-La-Or, T. Kulworawanichpong, And A. Oonsivilai

a) a)

b) b)
Fig. 6. Frequency response characteristics of the Fig. 7. Frequency response characteristics of the
experiment and the searching techniques with bi- experiment and the searching techniques with bi-
objective by weighted SSE (magnitude=0.25 and objective by weighted SSE (magnitude=0.00 and
phase=0.75): a) GA, b) ATS phase=1.00): a) GA, b) ATS

As can be seen, the results simulated by using the

parameters obtained from the bi-objective intelligent
search techniques are satisfactory and very much
better than those estimated by using a single
objective function because in frequency domain
magnitude and phase are equally essential
characteristics that cannot be ignored. Additionally,
according to comparison of SSE values, the GA
yields better results that fit to the experiments than
those obtained from the ATS which may be to notice
from SSE values.

Issue 3, Volume 3, March 2008 ISSN: 1790-5060

61
 WSEAS TRANSACTIONS on POWER SYSTEMS P. Pao-La-Or, T. Kulworawanichpong, And A. Oonsivilai

5 Conclusion in power system stability analyses, The Institute

This paper illustrates the bi-objective intelligent of Electrical and Electronics Engineers, Inc.,
approach to estimate parameters of a direct-axis 2003.
equivalent circuit transfer function of a synchronous [9] P.L. Dandeno, and P. Kunder, Stability
generator. As a result, magnitude and phase of its performance of 555 MVA turboalternators –
frequency response characteristics simulated from digital comparisons with system operating tests,
the proposed method satisfactorily fit to those IEEE Transactions on Power Apparatus and
obtained from the test data. However, utilizing the Systems, Vol.PAS-93, No.3, 1974, pp. 767-776.
bi-objective optimization is crucial due to the [10] T. Charuwat, and T. Kulworawanichpong,
difficulty of selecting weighted factors. It is Genetic based distribution service restoration
problem-dependent and system designers must with minimum average energy not supplied,
experience how to choose the weighted factors Proc. the 8th International Conference on
themselves in order to gain most advantages from Adaptive and Natural Computing Algorithms,
the parameter estimation proposed in this paper. Poland, 2007, pp. 230-239.
[11] D. Puangdownreong, T. Kulworawanichpong,
and S. Sujitjorn, Convergence analysis of
References: adaptive tabu search, ScienceAsia, Vol.30, No.1,
[1] H. Tsai, A. Keyhani, J. Demcko, and R.G. 2004, pp. 183-190.
Farmer, On-line synchronous machine parameter [12] X.F. Zha, and R.J. Howlett, Integrated
estimation from small disturbance operating intelligent systems for engineering design
data, IEEE Transactions on Energy Conversion, (chapter 12), IOS Press, Amsterdam, 2006.
Vol.10, No.1, 1995, pp. 25-35. [13] T. Kulworawanichpong, K-L. Areerak, K-N.
[2] IEEE Power Engineering Society, IEEE Std Areerak, P. Pao-la-or, P. Puangdownreong, and
115A-1987, IEEE Standard Procedures for S. Sujitjorn, Dynamic parameter identification of
obtaining synchronous machine parameters by induction motors using intelligent search
standstill frequency response testing, The techniques, Proc. the 24th IASTED International
Institute of Electrical and Electronics Engineers, Conference on Modelling, Identification, and
Inc., 1987. Control (MIC 2005), Austria, 2005, pp. 328-332.
[3] R. Escarela-Perez, T. Niewierowicz, and E. [14] P. M. Anderson, and A. A. Fouad, Power
Campero-Littlewood, Synchronous machine system control and stability, 2nd edition, Wiley,
parameters from frequency-response finite- 2003.
element simulations and genetic algorithms,
IEEE Transactions on Energy Conversion,
Vol.16, No.2, 2001, pp. 198-203.
[4] I. Kamwa, P. Viarouge, H. Le-Huy, and J.
Dickinson, A frequency-domain maximum
likelihood estimation of synchronous machine
high-order models using SSFR test data, IEEE
Transactions on Energy Conversion, Vol.7,
No.3, 1992, pp. 525-536.
[5] I. Kamwa, P. Viarouge, and J. Dickinson,
Identification of generalized models of
synchronous machines from time domain test,
IEE Proc. Conference, Vol.138, No.6, 1991, pp.
485-498.
[6] IEEE Power Engineering Society, IEEE Std 115-
1995, IEEE Guide: Test procedures for
synchronous machines, The Institute of
Electrical and Electronics Engineers, Inc., 1996.
[7] P. Kundur, Power system stability and control,
McGraw-Hill, 1994.
[8] IEEE Power Engineering Society, IEEE Std
1110-2002, IEEE Guide for synchronous
generator modeling practices and applications

Issue 3, Volume 3, March 2008 ISSN: 1790-5060

62