Research On Multi-Model Multi-Objective Evolution Algorithm With Integrated Double Spaces Learning

In the context of complex optimization,the information processing methods based on the von Neumann architecture has been challenged by the urgent need in efficiently solving the multi-modal,high-dimensional and large-scale problems.The computational intelligence method represented by evolutionary computing can provide an effective way to deal with multi-modal multi-objective optimization problems,and has become a frontier and hot research field in recent years.In order to solve several problems caused by the multi-objective evolutionary algorithm in solving MMOP,including the decision and objective space cannot be considered simultaneously,the global and local search is difficult to balance,the calculation efficiency is low and so on,the stochastic ranking-based multi-modal multi-objective differential evolutionary algorithm is proposed,the main research work is as follows:1.For traditional multi-objective evolutionary algorithm cannot optimize decision space and objective space simultaneously,global search and local search cannot be balanced when solving multi-modal multi-objective problems,a stochastic ranking-based multimodal multiobjective differential evolution algorithm(SRMMODE)is proposed.Firstly,under the framework of the fast non-dominated sorting,the crowding distances of solutions of critical layer in the decision space and objective space are measured separately,and the environmental selection is implemented through stochastic ranking,so as to achieve the diversity of population in two different spaces.Moreover,a multi-strategy differential evolution algorithm based on stochastic ranking is designed,which integrated the mutation strategy pool and ranking-based selection operators,in order to adjust the global and local search ability of the algorithm adaptively.The experimental results have shown that the comprehensive performance of SR-MMODE in the objective space and decision space is better than the other seven state-of-art algorithms.2.To further improve the computational efficiency of multi-modal multi-objective differential evolution algorithm,a multifactorial transfer learning-based multi-modal multi-objective evolution algorithm(MF-MMEA)is proposed.Firstly,in the early and middle stages of evolution,the multi-modal multi-objective differential evolution algorithm achieves the co-evolution of different optimization problems and the information transfered among different optimization problems through the assortative mating and the vertical cultural transfer under the framework of multifactorial optimization.Secondly,in the middle and later stages of evolution,the population has been divided into several subpopulations by skill factor,every subpopulation is optimized by the multi-strategy differential evolution algorithm based on stochastic ranking.The experimental results have shown that the computational efficiency of MF-MMEA is higher than the other state-of-art algorithms,and the performance in single problem is not significantly reduced.3.In order to verify the effectiveness of SRMMODE in practical problems,it is further applied to the cooperative path planning of multi UAVs.Firstly,by analyzing the cost of navigation and constraint conditions involved in the Cooperative Path Planning of multiple UAVs,a multi-objective optimization model is established by using the relevant mathematical tools.Moreover,SRMMODE is used to optimize this optimization model.The experimental results have shown that SRMMODE is effectivly in solving the problem of cooperative path planning of multi UAVs,and the necessity of multimodal properties in the multi objective problem discussed in this paper is proved.