Вы находитесь на странице: 1из 4

SYSTEM IDENTIFICATION USING TABU SEARCH

ALGORITHM
Aytekin BAI*, Nurhan KARABOA, Dervi KARABOA
Electronic Engineering Department, Engineering Faculty, Erciyes
University, 38039 KAYSER
*
e-mail: bagis@erciyes.edu.tr
Keywords: tabu search algorithm, system identification, ARMAX model
ABSTRACT: In system identification, the determination of model parameters is
an important problem. In this paper, the performance of tabu search algorithm is
measured for this problem. The results obtained from the simulation are found
very satisfactory.
I. INTRODUCTION
Using knowledge and observations, the construction of a mathematical
model for a dynamic system is called system identification [6-7]. Two elements
are important for system identification: the selection of a feasible model structure
and the determination of model parameters with the minimum error. If undesirable
inputs influence the system, it is quite diffucult to find the values of
parameters [1].
In this work, model structure is selected as ARMAX (autoregressive
moving average with exogenous variables) and the parameters of this model are
found using tabu search (TS) algorithm. ARMAX models equation is given as
t

k =1

k =0

k =0

y(n)= aky(n-k)+ bku(n-k)+ ckw(n-k)

(1)

where y(n) system output, u(n) system input, w(n) noise, a, b and c are unknown
parameters.
Model structure in this work is given as
y(n) = a y(n-1) + b u(n-1) + c w(n-1) + w(n)

(2)

In the following section, firstly the basic tabu search algorithm is briefly
described and then the application of tabu search to the determination of model
parameters is explained.
II. TS ALGORITHM AND ITS APPLICATION TO THE PROBLEM
TS algorithm is an efficient optimisation algorithm for difficult
optimisation problems [4-5]. The algorithm uses an initial solution which
represents the parameters to be found and new solutions are iteratively obtained
using tabu list and neighbourhood mechanism. Tabu list stores the information

about the past steps of search. This information is acquired by the recency and
frequency memories [2-5].
TS algorithm has a forbidding strategy [2-3]. New solution is the best
solution in a neighbourhood, which is not in tabu list. The use of tabu list is very
important because this list enables TS algorithm to escape from local minima
[4-5]. The flowchart of standard TS algorithm is given as follows [5]:
Simple Tabu Search Algorithm
{
Initial solution;
While termination criterion not reached
{ Neighbourhood generation
Evaluation of neighbourhood
Selection of best solution
Update tabu list
New solution }
}
The following evaluation function is used as objective function in the
determination of parameters.
n

e={( | yd(n) - y(n) |2 ) / n }(1/2)


k =1

where n is data number, yd and y are desired and actual outputs, respectively.
For the forbidding strategy of TS used in this work, two conditions are
used. These conditions are given as
Forbidding strategy
conditions

recency(i ) <= (rec).N

frequency(i ) >= ( freq ).(avgfreq)

(3)

where i is iteration number, N is element number in the tabu list, rec and freq are
recency and frequency factors ,respectively and avgfreq is the average change of
bits used to represent parameters.
In this work, the number of parameters to be found is nine. In order to
represent each parameter, five bits are employed. Therefore each solution is
represented using 45 bits. This is shown in Figure 1. The control parameter values
used are presented in Table 1.
Table 1. Control parameter values used
iteration (i)
tabu list size (N)
recency factor (rec)
frequency factor (freq)

=
=
=
=

100
9.0
0.1
2.0

[ 1 0 1 1 0 ...............................0 1 1 1 0]
[a1 a2 a3

b1 b2 b3

A solution (45 bits)

c1 c2 c3]

Tabu list (9 elements)

[a b c]

The model parameters

Figure 1. Representation of a solution in string form


TS uses a standart aspiration criteria to decide whether a solution is freed
or not. Thus, if a solution is better than the best solution found so far is freed. If
all solutions are tabu, then a least tabu solution is determined as the next solution.
input
w

W 0

U 0
-1
0

noise

10

20
30
data number
(a)

40

10

20
30
data number
(b)

40

output

10
y

-5
0

50

0
-10
0

10

20
30
data number
(c)

40

50

Figure 2. Input, noise and output for system


3.5
3
2.5
2
1.5
1
0.5
0
-0.5

13

17

21

-1

error (rms)

Figure 3. Model parameters and error for each iteration

25

50

The input, output and noise used in the simulation are given in Figure 2.
The model parameters (a, b, c) and error (rms) found for each iteration are shown
in Figure 3.
As the optimum parameters, a= 0.644988, b=1.120000 and c=2.567996
were produced using tabu search. The system output found by using these values
is given in Figure 4.
10
y

-10

data number

Figure 4. The system output


III. CONCLUSION
In this work, TS algorithm was used for a typical system identification
problem. It is concluded that TS is fast, flexible and very simple to implement.
The results found show that TS is very efficient for this kind of problem. TS can
be applied to more complex identification problems and its performance can be
compared with the performance of other algorithms such as genetic algorithm.
IV. REFERENCES
[1]A.Ba,Genetik
Algoritma
Kullanlarak
Gerekletirilen
Sistem
Kimliklendirme leminde aprazlama ve Mutasyon Operatrlerinin Etkisi,
V.Bilg.-Hab.Semp., Kasm 1998, Bursa, Trkiye, pp.222-225.
[2]D.Karaboa and A.Kaplan, Optimizing multivariable functions using tabu
search algorithm, The tenth Int.Symp. on Comp. And Inf.
Sciences,Oct.30,1995, Kuadas, Turkey, vol.II, pp.793-799.
[3]D.Karaboa,Design of Fuzzy Logic Controllers Using Tabu Search
Algorithm, 1996 Biennial Conf.of the North Ameraican Fuzzy
Inf.Pro.Society-NAFIPS, June 1996, U.S.A., pp.489-491.
[4]F.Glover, Tabu Search-Part I, ORSA Journal on Computing, Vol 1, No.3, pp.
190-206, 1989.
[5]F.Glover,A users guide to tabu search, Annals of Operations Research,
Vol.41, pp.3-28, 1993.
[6]K.Ogata, System Dynamics, Prentice Hall, New Jersey, 1992.
[7]L.Ljung,, System Identification: Theory for the User., Prentice Hall, 1987.

Вам также может понравиться