基于知识的分层遗传算法

A Knowledge-Based Hierachical Genetic Algorithm

传统遗传算法缺乏对进化过程知识的有效提取和利用,存在早熟收敛．在遗传算法的种群进化层上,引入文化算法的信度空间,提出一种具有知识引导功能的分层遗传算法．算法由底层种群进化层和上层知识进化层构成．结合遗传操作过程,提取4类知识并给出兵体定义．详细阐述了联系上下层的样本选取函数、知识更新函数和进化引导函数,并提出一种基于地势知识轮盘赌选择的新型个体替代策略．针对3组标准测试函数的仿真结果表明,4类知识在不同进化阶段对种群的影响程度不同．状况知识在进化早期起主导作用,规范知识和地势知识在某局部优势区域具有较强引导作用,历史知识引导搜索区域脱离局部较优点,从而有效避免早熟收敛,提高进化效率．

The knowledge about evolutionary process is not effectively abstracted and used in the genetic algorithm（GA） which is a premature convergence. A belief space in the culture algorithm was introduced to the genetic algorithm. A hierachical genetic algorithm with knowledge induction was proposed, which was composed of a lower population evolution layer and a upper knowledge evolution layer. Four kinds of knowledge were abstracted and defined through the analysis of evolutionary process. The sample choosing function, knowledge updating function and evolution inducting function were described in detail, which connect a lower layer and a upper layer. A novel substitute strategy to individuals based on roulette selection of topography knowledge was proposed. The simulation results of three benchmark functions indicate that four kinds of knowledge play the different role in different phases. The status knowledge influence evolution in the early phase. The limited optimum space is leading by normative and topography knowledge. The search space is escaped from optimum inducted by history knowledge so as to avoid premature convergence and advance the efficiency of evolution.