Research Journal of Applied Sciences, Engineering and Technology 4(2): 90-92 , 2012 ISSN: 2040-7467 © Maxwell Scientific Organization, 2012 Submitted: September 23, 2011 Accepted: October 24 , 2011 Published: January 15, 2012 Dynamic Multi-objective Optimization Algorithm Based On GEP and Virus Evolution Weihong Wang, Yanye Du, Qu Li and Zhaolin Fang College of Computer Science, Zhejiang University of Technology, China Abstract: Dynamic Multi-objective Optimization (DMO) is very popular nowadays. A new algorithm for DMO called Virus-GEP Dynamic based on Gene Expression Programming (GEP) and virus evolution is proposed. Experiments on two test problems have shown that the algorithm has better performance on convergence, diversity and the breadth of the distribution. Key words: Dynamic Multi-objective Optimization (DMO), Gene Expression Programming (GEP), virus evolution INTRODUCTION optimization algorithm NSGA-II (Deb et al., 2002) to make it suitable for DMO, while using GEP for encoding and virus evolution (Xian-bin et al., 2001; Shicheng et al., 2003) for evolution mechanism. Dynamic multi-objective optimization: Many systems need to consider dynamic scheduling problems, and these constraints are called dynamic constraints. Mathematical models abstracted from problems with multiple objectives and related with time factors are Dynamic Multi-objective Optimization (referred to as DMO) (Farina et al., 2002; Shang et al., 2007). Initialization: Initialization of the host population hostPop(t) and the virus population hostPop(t). Generate the subpopulation S by the host population P. Take genetic operations and selection operation to gain the new next host population hostPop(t+1). Then take virus-infection operation on each host individual, and the host individuals infected by the virus individual composesthe set U. Set the subpopulation S = hostPop (t+1). Then update the virus population. Generate the next host population by Pt and S: GEP: Gene Expression Programming (GEP) (Ferreira, 2001) was invented by Candida Ferreira. A GEP gene is the basic unit of a GEP genome and consists of head and tail parts. The gene is then mapped into an Expression Tree (ET) by following a width-first fashion and the ET is easy to be converted into a mathematical expression. • Virus evolutionary mechanism: As a major component of the biological immune system, virus system has many information processing mechanisms and functional features, and it has a great significance for further improvements in genetic algorithm, gene expression programming, and so on (Xian-bin et al., 2001). • New dynamic multi-objective optimization algorithm Virus-GEP dynamic coding: Each individual (chromosome) are composed of n genes, where n is equal to the number of decision variables of the multiobjective optimization problem. Each gene is responsible for one decision variable assignment. Suppose the genes {G1, G2, ..., Gn}, the corresponding terminal set for each gene Gi(i = 1, 2, ..., n) is Ti = {0, 1, 2, ..., 9, g1, g2, ..., gi-1} (i = 1, 2, ...., n), where figures 0, 1, ..., 9 show its location in the constants filed C. Generate the new population R = PcS, and do nondominated sorting of R to get the non-dominated fronts F!, F2, .... For Fi all, do sorting according to the CrowdedComparison Operator, and choose the N best individuals to compose the population P!. EXPERIMENTS AND ANALYSIS Convergence and diversity measures: For convergence measure, we use the measure of the objective space used in (Farina et al., 2002), which is shown in Eq (1). ef t 1 np np j 1 min S p, i (t ) X sol j i 1: nh (1) For diversity measure, we use the distribution of the optimal solutions obtained in the objective space to describe the diversity of the solutions. Flow of virus-GEP dynamic algorithm: The new algorithm is designed on the classic multi-objective Corresponding Author: Weihong Wang, College of Computer Science, Zhejiang University of Technology, China 90 Res. J. Appl. Sci. Eng. Technol., 4(2): 90-92, 2012 (a) Results of DBM[1] (b) Results of D-GEP Chaotic NSGA-II 1.5 1.0 -4.0 -4.5 -5.0 -0.5 lgeƒ lgeƒ 0.5 0 -1.0 -5.5 1.5 -6.0 -2.0 -6.5 -2.5 -3.0 -7.0 0 5 10 t 15 0 20 (c) Results of DBM[1] Fig. 1: Result of FDA 5 10 t 15 20 (d) Results of D-GEP Chaotic NSGA-II 2.0 2.0 1.5 f2 ƒ2 1.5 1.0 1.0 0.5 0.5 0 0 0 0.5 1.0 f1 1.5 0 2.0 1.0 f-1 1.5 2.0 (b) Results of D-GEP Chaotic NSGA-II 0 1.5 1.0 0.5 0 -0.5 -1.0 1.5 -2.0 -2.5 -3.0 -3.5 -1 lgeƒ lgeƒ (a) Results of DBM[1] 0.5 -2 -3 -4 0 2 4 6 t 8 10 -5 12 0 (c) Results of DBM[1] 2 4 6 t 8 10 (d) Results of D-GEP Chaotic NSGA-II Fig. 2: Result of FDA2 91 12 Res. J. Appl. Sci. Eng. Technol., 4(2): 90-92, 2012 Analysis of the experimental results: We use FDA1, FDA2 as test problems from (Farina et al., 2002) to study the dynamic multi-objective optimization problems. ACKNOWLEDGMENT The work is supported by the National Natural Science Foundation under grant 60873033, Natural Science Foundation of Zhejiang Province under grant R1090569 and Project of Department of Science and Technology of Zhejiang Province under grant 2009C31108. FDA1 test problem: Compare Fig. 1a with b, we find that, the new algorithm can find the optimal solutions in a wide range at each time step, that is, distribution of the obtained solutions is broader and uniform in the objective space. On the other hand, compare Fig. 1c with d, we can see that, convergence of the new algorithm is much better than DBM, the value of lg(ef) can keep change in the range of [-6.4, -4.3]. REFERENCES Deb, K., A. Pratap, S. Agarwal and T. Meyarivan, 2002. A fast and elitist multi-objective genetic algorithm: NSGA-II. IEEE T. Evol Comput., 6(2): 182-197. Farina, M., K. Deb and P. Amato, 2002. Dynamic multiobjective optimization problems: test cases, approximations and applications. IEEE T. Evol. Comput., 8(5): 425-442. Ferreira, C., 2001. Gene expression programming: A new adaptive algorithm for solving problems. Complex Syst., 13(2): 87-129. Shang, R., L. Jiao, M. Gong and M.A. wp, 2007. An immune clonal algorithm for dynamic multi-objective optimization. J. Software, 18(11): 2700-2711, (In Chinese). Shicheng, H., X. Xiaofei and D. Zhan, 2003. A Virus Evolutionary Genetic Algorithm for Large Product Structure Optimization Problem. J. Comp. Integrated Manufacturing Syst., 9(3): 202-205. Xian-bin, C., B. Wang, and X. Wang, 2001. A virus evolutional genetic algorithm [J]. Mini-micro Syst doi:cnki:sun:xxwx.0.2001-0.1-014. FDA2 test problem: Compare Fig. 2a with b, we can see that, uniformity of the new algorithm is poor, but to maintain a good diversity. On the other hand, compare Fig. 2c with d, convergence of the new algorithm is much better than DBM, the value of lg(ef) can keep change in the range of [-4.8, -0.7]. CONCLUSION We proposed a new algorithm Virus-GEP Dynamic based on GEP and virus evolution to solve DMO problems. Through the horizontal infection operation, this algorithm increases the diversity of the host population during the evolutional process, and make it easier to jump out of the local optimum, resulting in faster search for better solution. We took experiments on two test problems, and the results shown that the algorithm has better performance on convergence, diversity and the breadth of the distribution. 92