基于单基因变异算子的进化策略(μ+λ+к)-ES

Single-gene-mutation-based evolution strategies (μ+λ+к)-ES

中文关键词  进化策略  变异步长  变异算子  改进率

英文关键词  evolution strategies  mutation step-size  mutation operator  progress rate

中文摘要

      首先研究单基因变异进化策略(ES: evolution strategies)的变异步长控制，对变异步长与改进率的关系作理论分析，建立了基于单基因变异ES的变异步长控制策略．运用横向仿真方法比较分析了全基因变异算子和单基因变异算子对变异步长的适应性，表明单基因变异算子可以在变异步长比较大时、在较大变化范围内获得良好的收敛速度，可见对变异步长具有很强的适应性．在全局收敛性分析基础上，引入均匀变异算子作为附加算子，增强了全局收敛性，建立了一种基于单基因Gauss变异与均匀变异相结合的(μ+λ+к)-ES, 最后给出了仿真计算结果说明算法的有效性．

英文摘要

      The mutation step-size control for monogenic-mutation-based evolution strategies(ES) is investigated. After analyzing the relation between the step-size and the progress rate, we propose a step-size control scheme for the monogenicmutation-based ES. By transversal simulations, the adaptabilities of the hologenetic mutation operator and the monogenic mutation operator to the mutation step-size are analyzed by comparisons. Through simulation, it is shown that the monogenic mutation with a large mutation step-size may have a promising convergence-rate over a large varying range. This shows that the single-gene mutation is robust against the variation of mutation step-size. By analyzing the global convergence, we introduce a uniform mutation operator as an accessory operator to improve the global convergence performance. Thus, a new evolution strategy (μ+λ+к)-ES is established by combining the monogenic mutation Gauss mutation and the uniform mutation. Finally, simulation results are given to demonstrate the effectiveness of the algorithm.