基于启发式变异的改进演化规划算法

Improved Evolutionary Programming Algorithm Based on Heuristic Mutation

常用的演化规划算法(EP)的变异是基于固定的概率分布,鲁棒性较差.文中分析了变异算子对演化规划算法计算效率的影响,指出了Gauss变异、Cauchy变异和Lévy变异算子缺少启发式信息的不足,并据此设计了一种根据种群个体差异信息的启发式变异算子,用算子抽取的个体差异来更新变异步长,允许个体有机会在某些维数保持原状,只是进行部分维数上的变异.启发式变异算子能使演化规划算法更好地适应不同特点的连续优化问题,从总体上增强算法的鲁棒性.在求解多个Benchmark测试问题的数值实验中,基于启发式变异的改进演化规划算法比当前6种等概率分布演化规划算法有更快的收敛速度和更优的平均性能.

The common evolutionary programming EP） algorithms are of poor robustness because they perform the mutation based on a fixed probability distribution. In this paper, first, the influence of mutation operators on the computational efficiency of evolutionary programming algorithms is analyzed, and the essential drawback of Gauss, Cauchy and Levy mutation operators, namely the lack of heuristic information, is pointed out. Then, a heuristic mutation operator based on the differential information among individuals is designed, which uses the difference between two individuals to update the mutated variables and to provide chances for an individual to maintain its status quo in some dimensions. With the help of the proposed heuristic mutation operator, evolutionary programming algo- rithms can adapt to different continuous optimization problems and the algorithm robustness improves. Numerical experiments of several Benchmark problems demonstrate that the improved evolutionary programming algorithm based on heuristic mutation is of higher convergence speed and better average performance than six other evolutionary algorithms based on probability distribution.