一种两状态动态优化的自学习差异进化算法

A SELF-LEARNING DIFFERENTIAL EVOLUTION ALGORITHM WITH DUAL-STATE DYNAMIC OPTIMISATION

提出一种具有自学习机制的差异进化算法Se DE(Self-Learning Differential Evolution),提高在动态优化求解过程中群体适应环境动态变化的能力。采用重新评估特定个体的方式监测环境变化。通过群体向新状态历史最优解引导学习,将当代最优个体和两随机个体共同引导个体,保持群体多样性的同时加快算法收敛速度,降低环境的频繁变化对算法搜索能力的影响。通过6个动态函数测试了变化周期、函数维数对算法的影响,同时将新算法与两个同类算法比较,实验结果表明,自学习差异进化能更快地适应环境动态变化,获得更好的优化结果。

We proposed a differential evolution algorithm with self-learning mechanism for improving the capability of population in adapting to dynamic environmental changes in dynamically optimised solving process. By using the approach of re-evaluating specific individuals the algorithm monitors the environmental changes. Through population it guides the learning of the new state historical optimal solution,and guides the individuals by the contemporary optimal individual and the dual random individuals jointly; it keeps the diversity of the population while speeding up the convergence rate of the algorithm,and reduces the impact of frequent environmental changes on algorithm＇s search ability. We tested the influences of change period and function＇s dimension on the algorithm through 6 dynamic functions,and compared the new algorithm with two similar algorithms at the same time. Experimental result demonstrated that the self-learning differential algorithm could adapt to the dynamic environmental changes more rapidly and achieved better optimisation result.