MAS-Based Distributed Particle Swarm Optimization

Abstract

Without mutation and crossover operators, Particle Optimization Algorithm (PSO) is more liable to converge on a local minimum than genetic algorithm. Therefore, a distributed PSO based on multi agent system is proposed to overcome this shortcoming by importing mutation operators in genetic algorithm. In the evolution process, the agent swarms with different mutation operators run independently. Client agent swarms inform their individuals with highest fitness to the host agent swarm so as to replace its worst after certain generations of evolution. Therefore, the whole particle swarm enjoys a higher probability of finding the global optimum. Numerical results show the proposed method can find better solutions than other methods. Distributive in nature, it can run concurrently on many platforms scattered over local area network and even Internet for large-scale combinatorial combination problems.