Research On Multi-Objective Differential Evolution With Adaptation

Multi-objective optimization problem is a significant area of optimizations, there area lot of applications in reality have not only one object, even the objects are conflict toeach other. Sop can get one optimization solution, but because the conflict between MOP,the decider have to make a tradeoff in the solutions, which is the so-called Paretooptimization solution set. An amount of research works can prove that MOEA areavailable tools for MOP, draw a lot of attention for researchers.Differential evolution algorithm is one of the most important items in evolutionaryalgorithms; it has the characters that simple in principle, easy to operate, strongrobustness, so it is applied in optimizations broadly. The principle of differentialevolution algorithm is making use of the difference between individuals to generateperturbation, which can explore the research space. But if the selection of individual istoo stochastic, it may cause premature convergence. Beside of those, the setting ofparameters is significant, which can influence the performance of algorithm. Because ofthis, we make a deep research in the selection of mutation strategy and parameter controlwhen the differential evolution algorithms be applied to multi-objective optimizations,the main work of this paper include two aspects as follow:Operator selection and parameter setting are two important and active fields in theliterature of evolutionary algorithms. An adaptive multi-objective differential evolutionalgorithm with adaptive operator selection and online parameter control is proposed inthis section.Firstly, a clustering algorithm based on hierarchical aggregation is proposed toselect elitist individuals after the Pareto ranking that using the partial transitivity method;then update the using probability of each strategy according to the credit, and adjust thedistribution of parameters in the light of information carried by the reserved individualssimultaneously. Experimental results show validity and reliability in convergence anddistribution when compared to three state-of-the-art baseline approaches.The second, when solve MOP, the convergence and diversity must be emphasizedfor the performances of MOEA. But the traditional DE strategy can not balance thebreadth explorations and depth exploits, it makes the lace of guidance information andsufficient allocation, which can cause a bad convergence and diversity. For this we makeuse of local search to handle complex MOP. We test the algorithm in some test problemswith variable dependent; the results show the efficiency of the proposed algorithm.