一种自适应多策略差分进化算法及其应用

A self-adaptive differential evolution algorithm with multiple strategies and its application

针对差分进化算法由于固定参数设置而易早熟或陷入局部最优的问题,提出了一种自适应多策略差分进化算法(SMDE)。该方法以基本差分进化为框架,首先引入一个变异策略候选集合,一个缩放因子候选集合和一个交叉参数候选集合,然后在搜索过程中,以过去的搜索信息为基础,自适应地为下一时刻进化群体中的每个个体从候选集合中选择一组合适的变异策略和控制参数,以便在不同的进化时刻设置合适的变异策略和控制参数。对10个常用的标准测试函数进行优化计算,并与其他算法的结果进行了比较,实验结果表明,SMDE具有较好的搜索精度和更快的收敛速度。将SMDE用于化工过程动态系统不确定参数估计问题,实验结果表明该算法能较好地处理实际工程优化问题。

A self-adaptive differential evolution algorithm with multiple strategies（SMDE） was proposed to overcome premature or localized optimization of differential evolution（DE） as a result of fixed parameter settings. Based on basic framework of classical DE, the first step in SMDE was to create a candidate set of mutation strategy, scale factor（F） and crossover rate（CR）. In the followed searching process, mutation strategy, F and CR for each individual variable in next evolutionary generation were determined self-adaptively from the corresponding candidate set according to knowledge learnt from previous searches, so that proper mutation strategies and control parameters could be set at various evolution stages. Compared to other famous DE variants on optimizing 10 routine standard testing problems, SMDE had better search precision and faster convergence rate. Moreover, study on estimation of uncertain parameters in dynamic process systems of chemical engineering showed that SMDE could effectively solve engineering optimization challenges.