A Study Of Multi-swarm Particle Swarm Optimization Based On Information Sharing Mechanism

Particle Swarm Optimization algorithm(PSO)is a group of intelligent optimization algorithms,inspired by the phenomenon that birds how to find food.The algorithm has faster convergence speed and better global search ability.However,it is easy to fall into the local optimum due to the decrease of population diversity.The concept of multi-swarm is introduced into the particle swarm algorithm to enhance the population diversity of the algorithm and improve the convergence performance of the algorithm.The information between sub-swarms is not updated in time,which may decrease the search ability of the multiple swarms.It is a hot topic to strengthen the collaborative search ability between sub-swarms.In this paper,a multi-swarm particle swarm algorithm based on the periodic sharing mechanism and the knowledge billboard is proposed.The information between sub-swarms can be updated in time by introducing the cycle sharing mechanism.The knowledge billboard is used to record the information during the search process of the sub-swarms.The information in the knowledge billboard can be used to judge the search state of the sub-swarms and improve the convergence precision of sub-swarms.The main work in this study is as follow:1)The search information among the sub-swarms can not update timely.The multi-swarm particle swarm optimization based on the improved K-means and the periodic search mechanisms is proposed,which is called the IKMPSO method.Firstly,all particles are regarded as the root nodes of a tree.An optimal tree is generated by calculating the weights of all nodes and generating new nodes according to the weights.The appropriate leaf nodes of the optimal binary tree are chosen as the initial center point of the K-means cluster.The swarms are divided into several sub-swarms according to the improved K-means clustering method.The information between the sub-swarms is shared by a certain period of time by introducing the periodic sharing mechanism.In a certain update cycle,only one sub-swarm will search cooperatively under the guidance of its neighboring sub-swarms,and the remaining sub-swarms will search independently.The experimental results show that the proposed algorithm can greatly improve the convergence precision and convergence speed of the algorithm when solving the multi-peak test function.The searched information will be shared periodically among sub-swarms,which can enhance the cooperative search capabilities.2)In the IKMPSO algorithm,once a sub-swarm trapped in the local optimum,the sub-swarms adjacent it will be searched under the guidance of this local optimum.The diversity of the swarm will be decreased.The knowledge billboard sharing mechanism is introduced into the IKMPSO algorithm,which can increase the diversity of the swarm.A KBMPSO algorithm based on knowledge board sharing mechanism is proposed.The knowledge billboard is used to record the information that can be perceived in each sub-swarm search(the diversity of sub-swarms,the search ability of sub-swarms,and the optimal location and fitness values in sub-swarm search).Once a sub-swarm falls in the local optimum,the information recorded in the knowledge billboard is fed back to the sub-swarm.The sub-swarm adjusts its search direction according to the information,jumps out of the local optimum and searches in the global optimal direction.The convergence performance of KBMPSO algorithm is obviously improved compared with IKMPSO algorithm.