The Research Of Multi-Objective Particle Swarm Optimization

Along with the information technology development,the group intelligence algorithm agrees to give birth,this kind of algorithm can effectively overcome the classical algorithm cannot solve the model and has the advantages of simple operation and high solving efficiency.Therefore it becomes the research hotspot in the field of operations research.However,due to the randomness of the intelligent algorithm and the conflicts between each objective in the optimization process,the algorithm easily lead to premature convergence.In order to prevent the algorithm into local optimum and improve the robustness of the algorithm,this article put forward three different multi-objective particle swarm algorithms,including a multi-objective particle swarm optimization based on grid distance(GDMOPSO),a multi-objective particle swarm optimization based on global rank(GRMOPSO)and a multi-objective particle swarm optimization based on hybrid density learning(HDL-MOPSO).These three algorithms are based on the MOPSO algorithm,using the grid technology and Pareto dominant sorting principle to establish a new external archive control strategy and improve the learning samples.GDMOPSO algorithm updates and controls the external archive with the optimal grid distance strategy,and improves its learning samples.GRMOPSO algorithm is a new algorithm that replaces the original grid technology with the global ranking strategy where it establishes the relationship between the value of the objectives function and improves the external archive and learning samples with the acquisition strategy.HDL-MOPSO algorithm improves the global learning sample by combining global density and hybrid ranking strategy,so as to improve the robustness of the algorithm.We adopted a set of international standard benchmark functions to test the performance of the proposed improved algorithms compared with several classical multi-objective particle swarm optimization algorithms.The experimental results show that the algorithm can effectively avoid premature convergence and has good overall performance in generation distance,spacing and hyper-volume indexes.