Research On Particle Swarm Optimization For Solving Multi-Objective Optimization Problems

Many real-world problems belong to multi-objective optimization problems. Different with single-objective optimizations, competing goals give rise to a set of compromise solutions, generally denoted as Pareto-optimal. In the absence of preference information, none of the corresponding trade-offs can be said to be better than the others. What's more, these kinds of problems always have large and complex search space. Using traditional exact methods solve multi-objective problems may results high time-complex. Thus, designing new efficient optimization algorithms becomes imminent and realistic.The Particle Swarm Optimization (PSO) is a new Swarm Intelligence method based on the hypothesis that members of a population can profit from their past experience and the experiences of the other individual. Comparing with GA, this algorithm needs few parameters to set and can has faster convergence rates. But the research on PSO for MO problems is insufficient.We Compare PSO to Genetic Algorithms. And then the important strategies of GA for MO problems are discussed. The PAES and SPEA2 are also introduced. Based on the work above, two improved particle swarm algorithms to solve multi-objective problems are proposed. They use the information transfer schemes of PSO and the strategies of MOEA. Both of the two algorithms need archiving to maintain elitisms. The first algorithm adopts adapt grid archiving technique which is firstly used in PAES algorithm. Id order to obtaining more precise solutions, the information transfer of PSO is used in the whole population and the operators of crossover and mutation are added to the archiving. The second algorithm adopts PSO's information transfer schemes and environmental selection and matching selection strategy of SPEA2 algorithm to assure the population converge to the true Pareto front while keeping proper selection pressure.Using the benchmark continuous test problems, we compare the two algorithms with common MOPSO. Elitism, share and elimination strategy can be still valuable to PSO. Different information sources can also fast the convergent speed of the algorithm. The numerical experiment shows that the two improved algorithms can rapidly converge to the Pareto optimal front and spread widely along the front.