Surrogate-assisted Multi-objective Optimization Algorithm And Its Application Research

Currently,the application research of multi-objective optimization algorithm in the field of engineering optimization shows its high-efficiency computing capability and excellent algorithm performance.Although the multi-objective optimization algorithm has made great progress,the complexity of the engineering optimization design puts forward higher requirements on the performance of the multi-objective optimization algorithm.Since the multi-objective optimization algorithm requires a large number of function evaluations,it limits its application to expensive multi-objective optimization problems with high model complexity or low efficiency.Surrogate-assisted multi-objective optimization algorithms provide an effective method for solving such expensive multi-objective optimization problems.During the evolution,how to select effective correction points to improve the accuracy of the surrogate model is a prerequisite for the convergence of the algorithm.Therefore,surrogate-assisted multi-objective optimization algorithm is developed for expensive multi-objective optimization.Firstly,a fuzzy clustering based evolutionary algorithm called FCEA is proposed to optimize multi-objective optimization problems.During evolution of FCEA,the fuzzy c-means clustering method is firstly employed to discover the population distribution structure.Afterward,a membership based tournament selection(MBTS)strategy is designed to select similar parents for recombination by comparing membership value between solutions,which avoids the fitness evaluation of parents in the selection process.The use of MBTS significantly reduces the number of function evaluation.Comparison experiments show that FCEA outperforms four state of the art MOEAs on a set of test instances.The membership based tournament selection approach significantly contributes to the performance of FCEA.Then,a EI dominance principle based correction point selection approach is designed by means of utilizing Pareto dominance theory and the correction point selection strategy of the MOEA/D-EGO algorithm.And a Kriging model based expensive multi-objective optimization algorithm called FCEA-EGO is proposed in the paper.During the correction point selection process,a Par EI value for each candidate solution is obtained based on EI dominance and multiple EI values of each solution.A solution with lowest Par EI value is selected as the correction point for update the Kriging model.Comparison experiments show that FCEA-EGO outperforms MOEA/D-EGO,SMS-EGO and Par EGO on a set of test instances.Finally,the proposed expensive multi-objective optimization algorithm called FCEA-EGO is applied to gear reducer optimization problem for a light aircraft.Comparison experiment shows that FCEA-EGO outperforms MOEA/D-EGO on gear reducer optimization problem.It is feasible that solving practical engineering optimization problems with FCEA-EGO.