A Study And Application Of Multi-swarm Particle Swarm Algorithm Based On Shuffled Frog-leaping Algorithm And Mechanism Of Adding And Deleting

Multi-swarm particle swarm optimization has many advantages,such as the less parameters,the better ability of global searching and the better diversity of population.However,the ability of searching and sharing information is not fully used in the most of the improved algorithms.The communication between groups relies heavily on the reorganization of population,which leads to the lose of population diversity.However,compared with the particle swarm optimization algorithm,the shuffled frog-leaping algorithm has the better diversity of population.Therefore,in order to improve the diversity of population and the ability of global searching and local searching,this paper proposes an improved multi-swarm particle swarm optimization algorithm based on the shuffled frog-leaping algorithm(SFLA),which includes the multi-swarm strategy,the update strategy and the cooperation strategy.The multi-swarm strategy improves the ability of global searching by changing the way of population grouping.The update strategy improves the precision of poulation conbergence.The cooperation strategy improves the ability of population communication.In order to improve the diversity of population and the ability of jumping out of the local optimum,a mechanism is introduced to record the update state of the particles.Finally,the improved algorithm is applied to the feature selection of gene expression profiles.The main work of this article is as follows:1)In order to solve the instability of random classification,improve the ability of population communication and the diversity of population,this paper proposed a multi-swarm particle swarm optimization algorithm based on the mechanism of shuffled frog-leaping algorithm,which called multi-swarm leaping particles swarm optimization(MSLPSO).Firstly,a new classification strategy for particles is carried out in order to improve the ability of global searching.The particles is classified by the fitness values which are sorted in a descending order.Secondly,in the main group and the subgroup,the algorithm is updated with different strategies.The main group is responsible for convergence so as to improve the convergence ability of the algorithm and the subgroups is responsible for searching the whole space which improves the precision of poulation conbergence.Finally,the population uses a new information exchange mechanism.There is no direct communication between groups.The subgroups are only responsible for global searching.When a subgroup searches a better solution,the particle will communicate with the main group by replacing a certain particles in the main group to improve the ability of jumping out of local optimum.Compared with other PSO variants and SFLA,experimental results show that the algorithm based on SFLA has higher convergence accuracy and faster convergence speed.2)In order to improve the diversity of poplation,a new mechanism is proposed based on MSLPSO.The improved algorithm is applied to the feature selection method of gene expression profiles.First,an adding and deleting mechanism is proposed to record the update degree of particles.In this mechanism,each particle is given a periodic value.When the particle updates to a better state,the periodic value resets to 0,otherwise,the value increases by 1.If the particle reaches the predetermined periodic value,it will be deleted and regenerated a new particle to prevents the particle to get stuck and to improve the diversity of subgroups.This mechanism improves the diversity of population.Secondly,the improved algorithm is applied to the feature selection method in the gene expression data.Compared with other methods which using PSO variants,the results of the experiments on the gene expression data show that,based on the improved algorithm,the proposed method further improves the classification accuracy of the selected data sets.