The Research On Particle Swarm Optimization Algorithm To Solve Multi-Objective Optimization Problem

Evolutionary computation is greatly successful in handling complex and non-linear problems. Especially for multi-objective optimization problems, some attempts in this area have been made with significant results. The most representative algorithms are NSGA2 and SPEA2. Each run of these algorithms can maintain a solution set.Kennedy and Eberhart presented a new optimization method named Particle Swarm Optimization (PSO) in 1995, which is inspired by the flocking and swarm behavior of birds, insects, and fish schools. PSO is simple and performs efficiently, so many researchers have been attracted by this algorithm. Furthermore, it converges fast by moving each particle aimed at guides when it deals with single-objective optimization problem, and these features are useful in multi-objective optimization also.Using PSO to solve Multi-Objective Optimization problems, some research results have been presented. However, there are some shortcomings. First, the final solution can't spread to the Pareto frontier uniformly. Second, it is insufficient in convergence on many dimension problems. On the basis of the existing research work, this paper describes a new particle algorithm by employing a method of selecting global extremum and a new mutation operator to speed up the convergence. To improve the diversity, we put forward a density based PSO by using an archive. When the number of non-dominated solutions is bigger than the size of archive, the density method is used to pruning the archive. We also investigate the many dimension problems which can't converge to the true Pareto frontier. To overcoming this problem, we employ the decision-making incorporating Pareto ranking method. Two decision-making tables are adopted. One is generated randomly, and the other is invariable. By use of decision-making tables, a non-dominated solution is selected and all of its dominated solutions are eliminated. These make archive be close to the true Pareto frontier finally. The experimental results indicate that our new approach performs effectively on the many dimension problems, by comparing it with two of the state-of-art MOEAs and two well known MOPSO techniques on some well-established benchmarks.