递进多目标遗传算法

A Multi-Objective Genetic Algorithm Based on Escalating Strategy

在现有算法研究基础上,提出了一种递进多目标遗传算法,该方法每进化一定代数后以一定策略对群体进行重构,以提高算法对解空间的遍历性,从而较大程度上避免算法的早熟.该算法采用非劣解等级优先的选择方式复制后代,降低算法的时间复杂性;通过递进层次间对部分非劣解个体执行局部搜索,加快全局非劣解集的进化.采用递进算法与现有两种典型多目标遗传算法NSGA、MOGLS算法对一些典型优化问题进行对比分析,验证了算法求解多目标函数优化问题的有效性;通过调整算法递进层次与每层进化代数的参数设置,进一步研究了参数选取对算法性能的影响.

Multi-objective genetic algorithms are a kind of probabilistic optimization methods which concern with finding out a uniformly distributed non-inferior solution frontier to a given multi-objective optimization problem. A multi-objective genetic algorithm based on escalating strategy （EMGA） is proposed in this paper. The main idea of this escalating strategy is to re-generate the whole evolutionary population with some technology, which results in a new population significantly indifferent from the old one while inheriting the evolutionary information from the history. By this way, the performance on global convergence can be enhanced, and premature can be avoided simultaneously. A Pareto-ranking based selection strategy is used to reduce the computational expense of the algorithm, and a neighborhood search procedure is imposed on some selected Pareto solutions to accelerate the evolution process for reaching a global Pareto set with well distribution. Some typical multi-objective optimization test problems are taken to solve with EMGA, NSGA and M OGLS respectively to verify the effectiveness of the new algorithm. The details about how to select appropriate escalating parameters and their effect on the performance of EMGA are also investigated.