遗传算法中自适应伸缩搜索空间的方法

Self-adaptive search space expansion scheme for GA

在求解优化问题时,与进化策略和进化规划不同,遗传算法依赖于所给定的搜索空间。但对于大多数实际问题,并不知道最优解所在的区域,因而无法给出适当的搜索空间,大大影响了遗传算法的性能。针对这一问题提出了一种自适应伸缩搜索空间的方法,它包括扩展阶段和收缩阶段。前者能够快速找到一个包含全局最优解但较为粗糙的搜索空间,后者则不断地细化这个空间。文中方法可以从任意初始空间出发并很快获得较为精确的上下界。同时,当应用到动态环境时,也能够迅速地适应新的适应度曲面。仿真实验证明了其优越性能。

Unlike evaluation strategy (ES) and evaluation programing (EP) genetic algorithm (GA) strongly depends on the given search space for the optimal solution problem. The interval of existing optimal solution is unknown in most practical problem, then the suitable search space can not be given and the performance of GA are influence grently. A self-adaptive search space expansion scheme (SA-S^2ES) is proposed to cure this problem. It contains two steps: coarse-step and fine-step. The formen can quickly find a more coarse search interval including the overall situation optimal solution and the latter can refine this interval iteratively. The more precise lower and upper bounds can be obtained with this scheme from any initial search space. It is suited quickly to new suitability bend surface when dynamic environment is used. Simulation experiment results substantiate excellent performance of SA-S^2ES.