A Study Of Multi-swarm Particle Swarm Optimization Based On Chaotic Optimization And Its Applications

Multi-swarm particle swarm optimization(MPSO)is an optimization algorithm that combines local search with global search,where the population is divided into several sub-groups and each particle in the sub-group searches independently and shares information cooperatively.However,similar to the basic particle swarm optimization(PSO),the search process of MPSO tends to converge to the local optimal and lead to premature,since the speed is getting smaller and the particles may get to stagnate.As a non-linear phenomenon,the chaos has some characteristics such as regularity,ergodicity and randomness.By virtue of the advantages of chaotic search,the chaotic optimization can efficiently traverse the objective space with a small range,making the algorithm outperform traditional stochastic search algorithms.Therefore,this dissertation introduces a class of multi-swarm particle swarm optimization algorithms with chaotic optimization,and applies two hybrid algorithms to solve complex function optimization and gene selection of gene expression data.The mainly contributions of this dissertation are described as follows:(1)A hybrid multi-swarm particle swarm optimization algorithm(HMPSO-OCS)based on one-dimensional chaotic optimization and sequential communication is proposed to solve the optimization problem of single-objective continuous functions.Firstly,in order to enhance the information exchange among subgroups and improve the global search ability of the algorithm,a sort-based subgroup communication mechanism is designed to update the worst individuals of each subgroup.The individuals in each subgroups are sorted at first,and then the subgroups are sorted subsequently,leading the worst individuals in each subgroup have the opportunity to learn from the best individuals in the top-ranking subgroups,from which useful information can be obtained and updated.In addition,to enhance the local search ability of the algorithm,one-dimensional chaotic optimization is used periodically to optimize the global optimal individual in each subgroup.The experimental results show that the search performance of HMPSO-OCS is better than traditional algorithms including standard PSO,the improved PSO-based algorithms and MPSO.It achieves better search results on eight commonly used benchmark functions.(2)An improved binary multi-swarm particle swarm optimization(HMBPSO-EA)algorithm based on chaotic optimization and external queue is proposed for feature selection in gene expression data.With the ergodicity of chaotic optimization,HMBPSO-EA uses chaotic optimization to update the inertia weight of particles in order to balance the global and local search ability of particles.At the same time,to make full use of the information in each subgroup and assist the population evolution,an external queue is designed to store useful information of each subgroup for communication and information sharing among subgroups.The global optimal particle in each subgroup is the optimal individual of the current subgroup,and provides the optimal information of the subgroup to be stored in the external queue.Periodically extracting information from external queues to assist subgroups can make subgroups search more efficiently.Finally,HMBPSO-EA is applied in solving the problem of gene selection,and selecting the key characteristic genes in gene expression data to assist cancer diagnosis.The experimental results show that,compared with the traditional gene selection methods based on MBPSO and BPSO,the gene selection method based on HMBPSO-EA can find a gene subset with a higher classification performance with a better interpretability.