The Research On Large-scale Multi-objective Particle Swarm Optimization Algorithm Based On Excellent Particle Learning Mechanism

Multi-objective problems(MOPs)usually have multiple optimal solutions due to conflicting objectives.In addition,it is often difficult to distinguish between good and wrong solutions in the optimal solution set.Existing algorithms for solving multi-objective evolution problems(MOEAs)have shown promising results in solving problems in low-dimensional decision spaces.However,these MOEAs are significantly less optimized in high-dimensional decision spaces than in low-dimensional ones.In general,when faced with large-scale multiobjective optimization problems(LSMOPs)with more than 100 dimensions of decision variables,the search space with more than 100 dimensions poses a massive challenge to the search efficiency of general MOEAs.Furthermore,the number of local optima surrounded by extensive local regions can significantly increase.Therefore,it requires that the algorithm for solving large-scale MOPs has a good diversity preservation strategy and also needs to have a high convergence rate.The preference for learning swarm optimization algorithms in the current literature is to treat all particles in the population,so all particles use the same update method,but little attention is paid to the differences between particles.This paper proposes a level-based multistrategy learning swarm algorithm called LSLSO.In LSLSO,the optimization process is divided into a level-based search stage and a detailed search stage.First,particles are divided into four levels according to their fitness values.When the particles are at different levels in the population,LSLSO assigns different learning strategies to the particles of different levels to update the speed and position of the particles.Particles with better fitness focus on the development space.In contrast,particles with poor fitness will focus on exploring space.During the detailed search stage,particles of the same level learn from each other.Search in detail for small space particles in promising spaces that have already been explored.The theoretical analysis and the experimental results show that LSLSO has strong competitiveness in exploration and exploitation capabilities.In addition,this paper also proposes a self-exploroy competitive group optimization algorithm SECSO for large-scale multi-objective optimization.Particles evolve by exploring their neighborhood space and learning from other particles in the swarm,improving the algorithm’s diversity and convergence performance.The method has up to 10,000 decision variables on the LSMOP problems.Unlike most existing large-scale evolutionary algorithms that require extensive function evaluations,SECSO has the potential to find a set of well converged solutions when computational resources are scarce.It also has the potential to find more uniformly distributed Pareto optimal solutions when computing resources are sufficient.