一种基于超粒子引导的自适应知识迁移多任务差分进化算法

A super-particle guided multifactorial differential evolution algorithm with adaptive knowledge transfer

针对传统多任务优化算法(MTEA)存在负向知识迁移、迁移算子效率低下等问题,提出一种基于超粒子引导的自适应知识迁移的多任务差分进化算法(SAKT_MFDE).首先,通过任务之间的相似程度自适应地调节任务之间的交配概率,增大任务之间的正向迁移;其次,利用超粒子引导算法的搜索方向,进一步提升算法整体的优化效率;最后,通过多任务基准函数进行仿真实验来评价改进算法的寻优性能.实验结果表明,所提出算法可以有效规避任务之间的负向迁移,提高相似度较低的任务组的优化性能.

Aiming at the problems of negative knowledge transfer and low efficiency of the traditional multi-tasking optimization algorithm(MTEA),a multi-tasking differential evolution algorithm based on super-particle guided adaptive knowledge transfer(SAKT_MFDE)is proposed.Firstly,the algorithm adaptively adjusts the mating probability between tasks by the similarity degree between tasks to increase the forward migration between tasks.Secondly,the super-particle is used to guide the search direction of the algorithm,which further improves the overall optimization efficiency of the algorithm.The optimization performance of the improved algorithm is evaluated by the simulation of the multi-task benchmark function.The experimental results show that the proposed algorithm can effectively avoid the negative migration between tasks and improve the optimization performance of the task group with low similarity.