Study On The Global Search Strategies Of Multi-objective Particle Swarm Optimization

Particle swarm optimization (PSO) is one form of swarm intelligence,which is inspired by the behavior of bird ?ocks. The personal and the socialinformation are used to search the optimal solutions. Its high speed of conver-gence and relative simplicity make PSO draw more and more attention. PSOalgorithm has been successfully applied to many fields.The main di?culty in a multi-objective PSO (MOPSO) algorithm is howto identify the global best, because there is a set of compromise solutions ina multi-objective problem (MOP) rather than one best solution in a single-objective problem. The proportion of Pareto optimal solutions in a nondom-inated set grows with the number of objectives, which will a?ect on the per-formance of PSO algorithm. In addition, the fast convergence of PSO willresult in the local optimum. Premature convergence is caused by the rapidloss of diversity within the swarm. So when adopting PSO to solve MOPs, itis important to promote the diversity of population. In this dissertation, wepropose di?erent global search strategies to solve the aforementioned problems,and the main contributions of this dissertation can be summarized as follows:1. The basic concepts of MOP and the MOPSO algorithms are summa-rized, which is the preparation for the further research on the MOPSOalgorithms.2. A multi-swarm global search strategy based on fuzzy preference informa-tion is proposed, in which the prior preference of the user is consideredduring the optimization process. The preference weight value re?ects therelative importance of objective. The main swarm and assistant swarmare adopted. The preference information from the assistant swarm isconsidered by the information selection. The multi-swarm ensures thediversity and the convergence of the algorithm.3. An equilibrium selection of global search strategy based on preference or- der is proposed. This scheme can solve the selective pressure in the swarmcaused by the number of objectives increasing. The preference order(PO) is used instead of Pareto dominance to classify the nondominatedsolutions. PO can help reduce the number of points in a nondomiatedsolution set by retaining only those regarded as"best compromise", andthen"best compromise"as a global best updates the particle's velocity.The experimental results confirm that the proposed approach achievesbetter performance of the convergence and the su?cient diversity.4. An avoiding premature convergence of global search strategy based ontransposon operation is proposed, which can improve the diversity ofpopulation within the optimization process. The quality of the particleis changed by the transposon operation. In the optimization process, onlya few parameters are needed, which reduces the computational cost. Atthe same time, the transposon operation is ?exible and the convergence ofthe algorithm can be guaranteed. Simulations confirm that the proposedalgorithm is e?ective in maintaining the diversity of population.