基于Holt差分预测校正的动态多目标优化算法  

Dynamic multi-objective optimization algorithm based on Holt’s differential prediction correction

作　　者：刘志霖 康岚兰[1,2] 董文永 Liu Zhilin;Kang Lanlan;Dong Wenyong(School of Information Engineering,Jiangxi University of Science&Technology,Ganzhou Jiangxi 341000,China;School of Information Engineering,Gannan University of Science&Technology,Ganzhou Jiangxi 341000,China;School of Computing,Wuhan University,Wuhan 430072,China)

机构地区：[1]江西理工大学信息工程学院,江西赣州341000 [2]赣南科技学院信息工程学院,江西赣州341000 [3]武汉大学计算机学院,武汉430072

出　　处：《计算机应用研究》2024年第12期3701-3709,共9页Application Research of Computers

基　　金：国家自然科学基金资助项目(62166019);江西省自然科学基金资助项目(20232BAB202024);江西省研究生创新专项资金资助项目(YC2023-S660)。

摘　　要：为高效追踪动态多目标优化问题中随时间或环境变化而不断演变的Pareto前沿,提出了一种新的基于环境感知与预测校正的动态多目标优化算法(HD-DMOEA)。该算法包含三个主要策略:首先使用Wilcoxon符号秩检验对环境变化进行检测,并提出一种新的环境感知算子对环境变化强度进行判定。其次,构建Holt差分预测校正模型预测种群个体在下一个时间窗的位置,并在预测过程中根据参考点进行预测校正,以提高模型预测精度,加快算法寻优速度。另外,提出了一种新的变异方法,该方法根据环境变化强度引入不同的变异个体,以维持种群多样性,从而降低种群陷入局部最优的概率。为验证HD-DMOEA的有效性,将HD-DMOEA与五种最先进的预测算法分别在测试集FDA和dMOP上进行实验对比分析,实验结果表明,HD-DMOEA在搜索过程中能有效动态平衡种群的多样性和收敛性,实现对Pareto前沿的持续高效追踪,并且优于其他五种对比算法。To efficiently track the evolving Pareto frontier in dynamic multi-objective optimization problems over time or environment,this paper proposed a new dynamic multi-objective optimization algorithm,called HD-DMOEA,based on environment sen-sing and prediction correction.This algorithm incorporated three main strategies.Firstly,it employed the Wilcoxon signed rank test to detect environmental changes and introduced a new environment sensing operator to determine the intensity of these changes.Secondly,it constructed the Holt differential prediction correction model to predict the positions of population individuals in the next time window,making corrections based on a reference point during the prediction process to enhance prediction accuracy and accele-rate the search speed of the algorithm.Then,it introduced a new mutation method that injected different mutated individuals based on the intensity of environmental changes to maintain population diversity,thus reducing the likelihood of the population falling into local optima.To verify HD-DMOEA’s effectiveness,researchers conducted comparative analysis between HD-DMOEA and five advanced prediction algorithms on the FDA and dMOP test sets.The results demonstrate that HD-DMOEA effectively and dynamically balances the diversity and convergence of populations during the search process,continuously tracking the Pareto frontier and outperforming the other five comparative algorithms.

关 键 词：动态多目标优化 预测校正 环境感知 参考点 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]