731 Advances in Natural and Applied Sciences, 6(6): 731-740, 2012 ISSN 1995-0772

This is a refereed journal and all articles are professionally screened and reviewed

ORIGINAL ARTICLE

Modified Cat Swarm Optimization For Iir System Identification

1

Deivaseelan.A, PG Student, 2P.Babu, Associate Professor

K.S. Rangasamy College of Technology, Tiruchengode, Tamilnadu, India. Deivaseelan.A, PG Student, P.Babu, Associate Professor: Modified Cat Swarm Optimization For Iir System Identification

1,2

ABSTRACT Optimization problems are very important in signal processing. To the present, many optimization algorithms based on computational intelligence have been proposed, such as the Particle Swarm Optimization (PSO) and Cat Swarm Optimization (CSO). CSO is generated by observing the behavior of cats and composed of two submodels by simulating the behavior of cats. According to the experiments, the results reveal that CSO is superior to PSO. In this paper the IIR system identification task is formulated as an optimization problem and a recently introduced Modified Cat Swarm Optimization (MCSO), is used to develop a new population based learning rule for the model. Both actual and reduced order identification of few benchmarked IIR plants is carried out through simulation study. The results demonstrate superior identification performance of the new method compared to that achieved by particle swarm optimization (PSO) and Cat Swarm Optimization (CSO) based identification. Key words: System identification, IIR system, Particle Swarm Optimization, Cat Swarm Optimization, Modified Cat Swarm Optimization. Introduction Adaptive filters have become vastly popular in the area of digital signal processing. Adaptive direct modeling or system identification and adaptive inverse modeling or channel equalization find extensive applications in telecommunication, control system, instrumentation, power system engineering and geophysics (Zhenbin Gao; et al., 2008; Yamille del Valle, et al., 2008). Due to nonlinearity of the systems, System identification is a challenging and difficult optimization problem. Adaptive Infinite Impulse Response (IIR) systems are used in modeling real world systems because of their reduced number of coefficients and better performance over the Finite Impulse Response (FIR) filters (Sharbari Banerjee, Amitabha Sinha, 2009; Faten BEN ARFIA, 2009). Despite the fact that the digital IIR filter design is a well-researched area, major difficulties still exist in practice. This is because the error surface or the cost function of IIR filters is generally multimodal with respect to the filter coefficients. Thus, gradient-based algorithms can easily be stuck at local minima (D.J. Krusienski, W.K. Jenkins, 2004). This can be overcome by applying an optimization technique such as Particle Swarm Optimization (PSO). Due to its many advantages including its simplicity and easy implementation, the algorithm has been widely applied to optimize the Design of Adaptive Filters (Yamille del Valle, 2008). Computational intelligence is a hot research topic and many related algorithms have been proposed in recent years, such as the Genetic Algorithm (GA) (Davis, L., 1991; Pan, J.-S., et al., 1996; Iwasaki, N. and K. Yasuda, 2005), The Particle Swarm Optimization (PSO) (Kirkpatrick, S., et al., 1983; Huang, H.C., 2001), and the Simulated Annealing (SA) (Krusienski, D.J., Jenkins, W.K., 2005). Some of these optimization algorithms were developed based on swarm intelligence by simulating the intelligent behavior of animals. PSO finds the solutions by moving the particles in the solution space based on the balance of personal experience and best group experience. SA optimizes the solution based on the control of temperature for the acceptance or rejection of tuning the solution. PSO and CSO belong to the area of evolutionary algorithms. The idea of computational intelligence may come from observing the behavior of creatures. CSO was presented by studying the behavior of cats, and PSO was presented by examining the movements of flocking gulls. Through inspecting the behavior of the cat, Modified Cat Swarm Optimization (CSO) is proposed in this paper. Iir System Identification:

Corresponding Author: Deivaseelan.A, PG Student, K.S. Rangasamy College of Technology, Tiruchengode, Tamilnadu, India. E-mail: a.deivaseelan@gmail.com

732
Adv. in Nat. Appl. Sci., 6(6): 731-740, 2012

System identification is the mathematical modeling of an unknown system by monitoring its inputoutput data. This is achieved by varying the parameters of the developed model so that for a set of given inputs, its output match that of the system under consideration. For a plant whose behavior is not known, an adaptive system can be modeled and its parameters can be continuously adjusted using any adaptive algorithms. By the use of such adaptive algorithms, the required parameters can be obtained such that the output of the plant and the model are same for the same set of inputs, which is the goal of system identification. Such identification is a model of any arbitrary system. Most of the nonlinear systems are real and recursive in nature. Hence, models for real world systems are better represented as IIR systems. By doing so, the problem of system identification now becomes the problem of adaptive IIR filtering, for which different adaptive algorithms can be applied for adjusting the feed forward and feedback parameters of the recursive system. In the system identification configuration, the adaptive algorithm searches for the adaptive filter coefficients such that its input/output relationship matches closely to that of the unknown system is given by Fig:1.

Fig. 1: Block diagram of IIR system identification This can be written as difference equation of the form(16).

y(n) am (n) y(n m) bm (n) y(n m)

m1 m0

(1)

where x(n) and y(n) represent the kth input and output of the system. Also, n=1,2,3,.L and n=0,1,2,.M represent the coefficients of the IIR filter. Considering the block diagram of Fig.1,the output y(n) for input x(n) to the system is mixed with a noise signal v(n). The output of the system added with the noise gives the final system d(n). On the other hand, the output of the IIR filter in the modeled system for the same input x(n) has an output of (n). The difference of the output from the actual system with that of the modeled system gives the error e(n).This error is used by the adaptive algorithm to adjust the parameters of the IIR filter, and thus reduce the error in a number of iterations so as to exactly identify the actual system. Cat Swarm Optimization: There are about thirty different species of feline creatures, e.g., lion, tiger, leopard, cat etc, according to biological classification. Though many have different living environments, felines share similar behavior patterns. The hunting skill of felines is acquired through training although it is not innate for felines. For wild felines, this hunting skill ensures their food supply and survival of their species. Domestic cats also display the same natural hunting skill, and the strong curiosity of moving objects. Although all cats share this strong curiosity, they spend most of their time inactive. If you were to observe cats, you would notice thats cats spend most of their time resting, even when they are awake. Cats have a very high level of alertness. Seeking Mode: This sub mode is used to model the cat during a period of resting but being alert - looking around its environment for its next move. Seeking mode has four essential factors, which are designed as follows: Seeking Memory Pool (SMP): It is used to define the size of seeking memory of each cat, indication any points sort by the cat.

733
Adv. in Nat. Appl. Sci., 6(6): 731-740, 2012

Seeking Range of selected Dimension (SRD): It is used to declare the mutative ration for the selected dimensions. While in seeking mode, if a dimension is selected for mutation, the difference between the old and new values may not be out of range, the range defines by the SRD Counts of Dimension to Change (CDC): It is used to tell how many of the dimensions will be varied. All these factors play important roles in seeking mode. Self position consideration (SPC): It is a Boolean valued variable, and indicates whether the point at which the cat is already standing will be one of the candidate points to move to. SPC cannot influence the value of SMP. Seeking mode is described below. Step 1: Make j copies of the present position of catk, where j = SMP. If the value of SPC is true, let j = (SMP 1), then retain the present position as one of the candidates. Step 2: For each copy, according to CDC, randomly plus or minus SRD percents the present values and replace the old ones. Step 3: Calculate the fitness values (FS) of all candidate points. Step 4: If all FS are not exactly equal, calculate the selecting probability of each candidate point by equation (1), otherwise set all the selecting probability of each candidate point be 1. Step 5: Randomly pick the point to move to from the candidate points, and replace the position of catk.

Pi

| FSi FSb | , where 0 < i < j F S m ax F S m in

The goal of the fitness function is to find the minimum solution, where,

(2)

FSb = FSmax, otherwise FSb = FSmin. 3.2 Tracing Mode: Tracing mode is the sub-model for modeling the case of the cat in tracing targets. Once a cat goes into tracing mode, it moves according to its own velocities for each dimension. It can be follows: Step 1: Update the velocities for every dimension (Vid) according to equation (3). Step 2: Check if the velocities are in the range of maximum velocity. In case the new velocity is over-range, it is set equal to the limit. Step 3: Update the position of catk according to equation (4). Vid = W* Vid + C * r * (Pgd Xid ) (3)

where, W is the inertia weight, Pgd is the position of the cat, who has the best fitness value; Xid is the position of catk, C is a constant and r is a random value in the range of [0, 1]. Xid = Xid + Vid 3.3 Algorithm For Mcso: The MCSO algorithm reaches its optimal solution using two groups of cats, i.e. one group containing cats in seeking mode and other group containing cats in tracing mode. The two groups combine to solve the optimization problem. A mixture ratio (MR) is used which defines the ratio of number of cats in tracing mode to that of number of cats in seeking mode. 1. Randomly initialize the position of cats in D-dimensional space for the population, i.e. Xid representing position of ith cat in dth dimension. 2. Randomly initialize the velocity for cats, i.e. Vid. 3. Evaluate the fitness of each cat and store the position of the cat with best fitness as Pgm where m = 1, 2. , D. 4. According to MR, cats are randomly picked from the population and their flag is set to seeking mode, and for others the flag is set to tracing mode. 5. If the flag of ith cat is seeking mode, apply the cat to the seeking mode process, otherwise apply it to the tracing mode process. The steps of the corresponding modes are followed. 6. Evaluate the fitness of each cat and store the position of the cat with best fitness as Plm where m = 1,2,. . . ,D. 7. Compare the fitness of Pg and Pl, and update Pg. (4)

734
Adv. in Nat. Appl. Sci., 6(6): 731-740, 2012

8. Check the termination condition, if satisfied, terminate the program. Otherwise repeat steps 47. 9. Calculate the velocity function vid of update new velocity and otherwise repeat steps 27. The term Vid is limited to the range Vmax. If the velocity violates this limit, it will be set at its proper limit. Changing velocity enables every cat to search around its individual best position and global best position. Based on the updated velocities, each cat changes its position. In addition to this maximum velocity is also limited using update new velocity. As velocity update leads to acceleration of cats. However, too small an acceleration may lead to slow convergence, whereas too large an acceleration drives the cats towards infinity. 4. Simulation Results: In this section, firstly the performance of the PSO algorithm is compared with that of the CSO algorithm and its improved version MCSO on a set of co-efficient; and secondly MCSO, CSO, and PSO algorithms are applied to the design of full and reduced order digital IIR filters for the purpose of system identification. 4.1 First Example: For First Example, transfer functions of unknown plant and filter are given in case Johnson, C.R. Jr., (1984), where [a0,a1,b1,b2] the parameters to be are founded their optimum values are [0.05,-0.4, -1.1314 and 0.25]. The coefficients are randomly initialized within the periphery of the possible solution and the maximum velocity and position are also restricted to maximum velocity and position which is the maximum value of the actual coefficients of the plant. The transfer functions and their implementation in two different models for estimated value are shown in Table 1. These IIR systems are modeled using two different models, one having the same order as the actual system and second having less order than that of the actual system. These reduced order cases pose challenge to the optimization algorithm in that they produce multimodal error surface which has multiple minima. In all cases, as the number of coefficients decreases, the degree of freedom reduces and it becomes more difficult to identify the actual system.

Fig. 2: Block diagram of MCSO based IIR system identification

Table 1: Parameter Estimation of II order IIR Filter Parameter1 Actual value a0 0.05 a1 -0.4 b1 1.1314 b2 -0.25 Computational time(in seconds) Transfer function1 Full Order L M Model Reduced Order L M Model Estimated value PSO 0.1048 -0.6131 0.7626 -0.1624 30.738748

CSO 0.0142 -0.2949 0.7325 -0.0856 39.358608

MCSO 0.0321 -0.4739 1.0162 -0.2918 41.375335

Hs (z)
2 1

0.050.4z1 11.1314z1 0.25z2

a 0 a1 z 1 1 b1 z 1 b 2 z 2

H s (z)
1 0

(z)

a0 1 b1 z 1

735
Adv. in Nat. Appl. Sci., 6(6): 731-740, 2012

Fig. 3: Flow chart for MCSO 4.2 Second Example: The transfer functions of unknown plant and filters are given in case Majhi, B., & Panda, G. (2010).
Table 2: Parameter Estimation of III order IIR Filter Parameter2 Actual value Estimated value PSO a0 -0.2 -0.2788 a1 -0.4 -0.4302 a2 0.5 0.5265 b1 0.6 0.6595 b2 -0.25 -0.5340 b3 0.2 0.5156 Computational 30.471340 time(in seconds)

CSO -0.1734 -0.4207 0.4426 0.4386 -0.0381 0.1769 39.233766

MCSO -0.1528 -0.2042 0.4005 0.4714 -0.0544 0.1830 41.623385

736
Adv. in Nat. Appl. Sci., 6(6): 731-740, 2012

Case 2: Transfer function2

H s ( z)
Full Order L M Model Reduced Order L M Model 3 2

0.2 0.4 z 1 0.5z 2 1 0.6z 1 0.25z 2 0.2z 3

a0 a1 z 1 a2 z 2 1 b1 z 1 b2 z 2 b3 z 3 a 0 a1 z 1 1 b1 z 1 b 2 z 2

H s ( z)
2 1

H s (z)

4.3 Third Example:

Table 3: Parameter Estimation of IV order IIR Filter Parameter3 Actual value a0 1 a1 -0.9 a2 0.81 a3 -0.729 b1 0.04 b2 -0.2775 b3 0.2101 b4 -0.14 Computational time(in seconds) Case 3: Transfer function3 Estimated value PSO 0.8784 -0.9427 0.2227 -0.6518 -0.0695 -0.1258 0.1139 -0.2392 30.428329

CSO 0.7795 -0.8689 0.9253 -0.4416 -0.1543 -0.1091 0.0715 -0.5314 39.071058

MCSO 0.7947 -0.5449 0.6862 -0.8936 -0.0548 -0.3037 0.0462 -0.1955 41.316061

Hs (z)=
Full Order L M Model 4 3

-0.2-0.9z-1 +0.81z-2 -0.729z-3 1-0.04z-1 +0.2775z-2 +0.2101z-3 +0.14z-4

a0 a1 z 1 a2 z 2 a3 z 3 H s ( z) 1 b1 z 1 b2 z 2 b3 z 3 b4 z 4
3 2

Reduced Order

L M Model

H s (z)

a 0 a1 z 1 a 2 z 2 1 b1 z 1 b2 z 2 b3 z 3

4.4 Forth Example:

Table 4: Parameter Estimation of V order IIR Filter Parameter Actual value4 Estimated value PSO a0 0.1084 0.1613 a1 0.5419 0.9118 a2 1.0837 0.3145 a3 1.0837 0.1338 a4 0.5419 0.0035 a5 0.1084 0.1223 b1 -0.9853 -0.1045 b2 -0.9738 -0.4999 b3 -0.3864 -0.4562 b4 -0.1112 -0.1314 b5 -0.0113 -0.5548 Computational time(in seconds) 30.315452

CSO 0.4818 0.5138 0.6655 0.0398 0.2178 0.3753 -0.2324 -0.1595 -0.5677 -0.0580 -0.0195 39.082723

MCSO 0.0979 0.3751 0.9101 0.9898 0.5990 0.2984 -1.0817 -0.5960 -0.4262 -0.3559 -0.0571 41.061856

737
Adv. in Nat. Appl. Sci., 6(6): 731-740, 2012
Case 4: Transfer function 4 Full Order L M Model Reduced Order L M Model

Hs ( z)
5 5

0.1084 0.5419z1 1.0837z2 1.0837z3 0.5419z4 0.1084z5 1 0.9853z1 0.9738z2 0.3864z3 0.1112z4 0.0113z5
a 0 a1 z 1 a 2 z 2 a 3 z 3 a 4 z 4 a 5 z 5 1 b1 z 1 b 2 z 2 b3 z 3 b 4 z 4 b5 z 5

H s (z)
4 4

a 0 a1 z 1 a 2 z 2 a 3 z 3 a 4 z 4 H s (z) 1 b1 z 1 b 2 z 2 b 3 z 3 b 4 z 4

The estimated parameters listed in table [1,2,3,4] shows that MCSO based learning rule is able to estimate the coefficients better than PSO,CSO and MCSO. The MCSO requires a higher computation time in comparison to PSO and CSO. The MCSO performs best in identifying a system using a reduced order model in case[1,2,3,4]. The simulation results shows that higher convergence speed of MCSO makes it better choice for all[2st ,3rd,4th,5th ] order system. The fig[4,5,6,7] indicates that the MCSO converges faster than the PSO but performs almost equally like CSO. It almost PSO, CSO MCSO are equal but convergence speed is varied. The number of cats as well as the number of iteration are varied for maintain thetotal number of fitness function evaluations. The simulation using 50 and 100 cats are carried out.

Fig. 4: convergence characteristic (2nd order filter)

738
Adv. in Nat. Appl. Sci., 6(6): 731-740, 2012

Fig. 5: Convergence characteristic (3rd order filter)

Fig. 6: Convergence Characteristic (4th order filter)

739
Adv. in Nat. Appl. Sci., 6(6): 731-740, 2012

Fig. 7: Convergence characteristic (5th order filter) Conclusion: The natural world conceals many characteristics of different creatures, and all of them have some unique behaviors or features to keep them survive. In this paper, we present a new algorithm, Modified Cat Swarm Optimization, through modeling the behaviors of cat to solve the optimization problems. The experimental results indicate that MCSO can better improve the performance on finding the global best solutions. To compare with PSO, CSO algorithms, the performance assessment of the MCSO, in identifying an unknown system with a reduced order model, is also carried out, i.e. the maximum velocities, in all iterations. And it can locate the global best solution much faster than PSO, CSO algorithm. References Johnson, C.R. Jr., 1984. Adaptive iir filtering: Current results and open issues. IEEE Transactions on Information Theory, 30: 237-250. Majhi, B., & G. Panda, 2010. Development of efficient identification scheme for nonlinear dynamic systems using swarm intelligence techniques. Expert Systems with Applications, 37: 556-566. Ng, S., S. Leung, C. Chung, A. Luk, & W. Lau, 1996. The genetic search approach: A new learning algorithm for adaptive iir filtering. IEEE Transactions on Signal Processing, 13: 38-46. Regalia, P.A., 1992. Stable and efficient lattice algorithms for adaptive iir filtering. IEEE Transactions on Signal Processing, 40. Zhenbin Gao, Xiangye Zeng, Jingyi Wang, Jianfei Liu, 2008. FPGA implementation of adaptive IIR filters with particle swarm optimization algorithm ICCS-2008; 11th IEEE International Conference on Communications Systems held on 19 to 21 Nov 2008 in China; Page(s): 1364-1367. Yamille del Valle, Ganesh Kumar Venayagamoorthy, Salman Mohagheghi, Jean-Carlos Hernandez, and Ronald G. Harley, 2008. Particle Swarm Optimization: Basic Concepts, Variants and Applications in Power Systems IEEE Transactions on Evolutionary Computation, 12(2): 171-195. Sharbari Banerjee, Amitabha Sinha, 2009. Performance Analysis of Different DSP Algorithms on Advanced Microcontroller and FPGA ACTEA 2009; IEEE International conference held on July 15-17, 2009 Zouk Mosbeh, Lebanon; Page(s): 609-613. Faten BEN ARFIA, Mohamed Ben Messaoud, Mohamed Abid, 2009. Nonlinear adaptive filters based on PSO Leonardo Journal of Sciences, Issue 14, Jan-Jun 2009. Krusienski, D.J., W.K. Jenkins, 2004. PSO for Adaptive IIR Filter IEEE Transactions on Signal processing. Goldberg, D.E., 1989. Genetic Algorithm in Search, Optimization and Machine Learning, AddisonWesleyPublishing Company. Davis, L., 1991. Handbook of Genetic Algorithms, Van Nostrand Reinhold, New York.

740
Adv. in Nat. Appl. Sci., 6(6): 731-740, 2012

Pan, J.-S., F.R. McInnes and M.A. Jack, 1996. Application of parallel genetic algorithm and property of multiple global optima to VQ codevector index assignment, Electronics Letters, 32(4): 296-297. Iwasaki, N. and K. Yasuda, 2005. Adaptive particle swarm optimization using velocity feedback, International Journal of Innovative Computing, Information & Control, 1: 3. Kirkpatrick, S., Jr. C.D. Gelatt and M.P. Vecchi, 1983. Optimization by simulated annealing, Science, 671-680. Huang, H.C., J.S. Pan, Z.M. Lu, S.H. Sun and H.M. Hang, 2001. Vector quantization based on generic simulated annealing, Signal Processing, 81(7): 1513-1523. Krusienski, D.J., W.K. Jenkins, 2005. Design and performance of adaptive systems based on structured stochastic optimization strategies. IEEE Circuits and Systems Magazine, 5: 8-20.