Self-Adaptive Local Search For Promoting The Performance Of MOEA/D

Traditional optimization methods have many limitations in solving optimization problems that characterized by highly complexity, non-convexity, multi-extremum and other features. Evolutionary algorithm(EA) is a random searching algorithm with good robustness and universality based on Darwin's theory of evolutionary. It can effectively solve problems which cannot be tackled with traditional mathematical optimization methods. In recent years, Multi-objective Evolutionary Algorithm(MOEA) attracts great attention due to its capability of solving complex optimization problems with multiple targets, has become a hot topic in computing intelligence.Multi-objective Evolutionary Algorithm Based on Decomposition(MOEA/D) is a new evolutionary algorithm that can decompose a multi-objective optimization problem into a group of single-objective sub-problems. From the perspective of the elite individuals selection mechanism, MOEA/D is different from the algorithm based on Pareto Optimal Theory(e.g. NSGA-II) and other algorithms based on performance indicator(e.g. IBEAs). Each sub-problem is essentially a scalar function,therefore the evaluation of an individual ' s superiority becomes a direct comparison of fitness function value of individuals, and retaining of the elite individuals in population becomes simple and efficient.In this paper, a new version of Multi-objective Evolutionary Algorithm Based on Decomposition with adaptive local search strategy is proposed.The most appropriate operator for the question and the current evolution phase can be selected through evaluating the performance of recombination methods during the execution of the algorithm.In addition, a self-adaptive local search strategy is proposed in order to avoid premature solutions and keep individuals away from trapped in local optimal by adjusting the search space of sub-problems adaptively through evaluating the improvement of sub-problem solutions during some generation. This strategy improves the convergence ability of the algorithm and keep solutions diverse as well.Overall, the work of this paper includes the following sections:1)An adaptive selection strategy of crossover operator is proposed. Several typical crossover strategies are choosed as candidate operators. The most suitable operator for the problems in current stage of the execution of the algorithm can be selected adaptively according to the performance of each crossover operator.Thisstrategy can improve global optimization ability of the algorithm remarkably.2) An adaptive local search strategy is also proposed to adjust the the search scope of each sub-problem based on the judgement that the sub-problem has trapped in local optimal or not by evaluating the improvement of sub-problem 's solution during a stage of evolution.This strategy prevents solution set from premature convergence, and improve the convergence efficiency of the algorithm.3) A new version of MOEA/D with an adaptive local search strategy included in is proposed,and named as MOEA/D-SLS.Then the algorithm is tested on the ZDT and DTLZ series of multi-objective test problems and compared with two famous evolutionary multi-objective optimization algorithms. Some discussion on the advantages and disadvantages of each algorithm referred in solving multi-objective optimization problem from the perspective of the performance indicator follows.At the end of this paper, we present some investigation worthing further studying,and some conclusion is given.