一种基于混合交叉的差分进化算法

Differential evolution algorithm based on hybrid crossover

针对差分进化算法中指数交叉方式的研究盲点,提出一种混合交叉及扩展的带系数混合交叉算法.混合交叉能改善指数交叉中尝试向量从变异向量选择的分量期望值太小和尝试向量候选集太小的问题,还能保持指数交叉善于处理含有相邻非独立变量以及二项式交叉善于处理不相邻非独立变量的优点.将这两种交叉方式与一种带档案集的自适应参数差分进化算法相结合,并采用无约束实参函数优化测试集验证了该混合交叉有效性与优越性.数值实验结果表明该混合交叉的效果优于单纯的二项式交叉和指数交叉.

Aiming at the blind spot of exponential crossover in differential evolution algorithm,a hybrid crossover and an extended hybrid crossover method with a coefficient was proposed.The hybrid crossovers overcame the exponential crossover’s disadvantages that the expected value of the component selected from the mutation vector and the trial vector candidate set was too small.They could also maintain the exponential crossover’s advantage of dealing with adjacent non-independent variables and the binomial crossover’s advantage of dealing with non-adjacent independent variables.The two crossovers were combined with an adaptive differential evolution variant with an external archive.Validity and superiority of the proposed hybrid crossovers were verified by an unconstrained real parameter optimization benchmark.Numerical experiment results show that the proposed hybrid crossovers perform better than exponential crossover and binomial crossover.