Research On Improved Method Of Multi-objective Particle Swarm Optimization Algorith

In recent years,how to solve complex multi-objective optimization problems(MOPs)in the engineering field has attracted wide attention,and many optimization algorithms based on biological population behavior have been proposed.Among them,Particle swarm optimization(PSO)based on swarm intelligence has great development potential in the field of multi-objective optimization due to its characteristics of fewer parameters,fast convergence speed and easy implementation.However,if PSO is used to solve MOPs,some key problems need to be considered,such as how to balance the optimization ability and convergence ability of the algorithm,how to select global learning samples,how to maintain the external archive scale and so on.These problems bring severe challenges for PSO to solve MOPs.Therefore,this paper proposes several improved algorithms,as follows:1)In order to improve the performance of the algorithm and reduce the probability that the algorithm falls into local convergence during the solution process,a multiobjective particle swarm optimization based on ideal distance(IDMOPSO)is proposed.In IDMOPSO,adaptive parameters are introduced to dynamically adjust particle’s velocity in the population,in order to balance the optimization and convergence capabilities of the algorithm.Then,adaptive mesh and ideal distance are used to optimize the selection method of global learning samples and improve the control strategy of external archive size.In addition,in order to prevent IDMOPSO from falling into local optimization,cosine factors are introduced in the process of development and exploration to vary the position of particles.In the early stage of this mutation stage,particles in IDMOPSO belong to the development process,so the variation range is small,while in the late stage,the variation range is increased due to the local optimization of the particle.Finally,IDMOPSO,several popular multi-objective particle swarm optimizers(MOPSOs)and other multi-objective evolutionary algorithms(MOEAs)were simulated on ZDT,DTLZ and UF series test functions.The experimental results show that IDMOPSO has the better convergence,diversity,and excellent solution ability compared to the other algorithms.2)In order to enhance the ability of selecting and storing excellent particles in the algorithm,a novel multi-objective particle swarm optimization based on ranking and cyclic distance(RCDMOPSO)is proposed.In RCDMOPSO,a global proportion ranking method is introduced,which is different from non-dominated ranking under Pareto framework,and a diversity evaluation strategy based on cycle distance is proposed.A novel strategy for selecting global learning samples and maintaining external archives is designed by combining global proportion ranking with cyclic distance.Finally,RCDMOPSO,several popular MOPSOs and other MOEAs were simulated on ZDT,DTLZ and UF series test functions.The experimental results show that RCDMOPSO has the better convergence and diversity.3)In order to improve the quality of non-inferior solutions in external archive and increase the selection pressure of learning samples,a multi-objective particle swarm optimization based on double decision and fast stratification(DDFSMOPSO)is proposed.In DDFSMOPSO,the double decision of the combination of crowding distance and absolute distance is used to maintain the external archive scale,so that excellent particles can be easily retained and developed in the subsequent evolution process.At the same time,global learning samples are selected from the external archive using a fast stratification strategy,which are used to lead the evolution of particles in the population and promote the particles in the population to move to the real Pareto frontier.In addition,a speed update method based on dual strategy is designed to balance the local and global search capabilities of the algorithm.Simulation results show that DDFSMOPSO can be used as an effective algorithm for solving MOPs.