一种带有三重选择机制的多种群多策略差分进化算法  

Multi-population multi-strategy differential evolution algorithm with triple selection mechanism

作　　者：宋晓宇[1] 李敏[1] 赵明[1] Song Xiaoyu;Li Min;Zhao Ming(School of Computer Science&Engineering,Shenyang Jianzhu University,Shenyang 110168,China)

机构地区：[1]沈阳建筑大学计算机科学与工程学院,沈阳110168

出　　处：《计算机应用研究》2025年第3期795-803,共9页Application Research of Computers

基　　金：辽宁省教育厅重点科研资助项目(lnzd202004);辽宁省科技厅自然科学基金计划资助项目(2023-MS-222)。

摘　　要：针对差分进化算法(differential evolution,DE)在寻优过程中易陷入局部最优以及求解精度不高的问题,提出一种带有三重选择机制的多种群多策略差分进化算法(TSMDE)。该算法采用分层种群结构,利用适应度值将种群划分为三个子种群,且子种群的大小随迭代动态调整。同时,采用五个改进的突变策略以及不同的参数自适应方式,以满足个体在不同进化阶段的开发与探索需求。为了充分发挥多种群的优势,提出一种高效的信息共享机制——三重选择机制。各子种群先根据不同模式选择执行突变的个体,然后该个体根据自身进化状态选择合适的突变策略,最后判断出该个体处于停滞状态后从两个外部存档中选择一个候选解进行替换,最终通过三重选择机制引导整个种群的进化进程。最后,将TSMDE与13个先进的差分进化(DE)变体进行对比,以验证TSMDE的有效性。在CEC2014测试集中的30个基准函数上的实验结果表明,该算法在求解精度、避免陷入局部最优等方面的能力优于或比得上这13个先进算法。To address the issue of the differential evolution(DE)algorithm falling into local optima and achieving low solution accuracy during the optimization process,this paper introduced a multi-population multi-strategy differential evolution algorithm(TSMDE)with a triple selection mechanism.The algorithm adopted a hierarchical population structure,dividing it into three sub-populations based on fitness values,and dynamically adjusted the size of each sub-population as iterations progress.Meanwhile,it applied five improved mutation strategies as well as different parameter adaptive approaches to balance exploitation and exploration for individuals at different evolutionary stages.In order to fully utilize the advantages of multiple populations,it implemented an efficient information-sharing process via the triple selection mechanism.In this mechanism,each sub-population first selected an individual to mutate using different modes.Then,the individual chose a suitable mutation strategy based on its evolutionary state,and finally,it selected a candidate solution from two external archives to replace the target vector when stagnation was detected.This guided the evolutionary process of the entire population.Finally,it compared TSMDE with 13 advanced DE variants to validate its effectiveness.Experimental results on 30 benchmark functions from the CEC2014 test set demonstrate that the proposed algorithm achieves better or comparable performance in both solution accuracy and avoiding local optima compared to the 13 advanced algorithms.

关 键 词：差分进化 分层种群 多策略 三重选择机制 参数自适应 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]