基于混合引导策略的偏好多目标进化算法  

Preference multi-objective evolutionary algorithm based on hybrid guidance strategy

作　　者：王沛东 祝园园 孙希霞 WANG Peidong;ZHU Yuanyuan;SUN Xixia(College of Internet of Things,Nanjing University of Posts and Telecommunications,Nanjing 210003,China)

机构地区：[1]南京邮电大学物联网学院,南京210003

出　　处：《智能计算机与应用》2025年第3期125-132,共8页Intelligent Computer and Applications

摘　　要：针对传统偏好多目标进化算法存在的算法性能受偏好点位置影响,不易于控制偏好解集大小以及收敛速度较慢等问题,提出了一种基于角度和距离混合引导策略的偏好多目标进化算法。首先,设计了一种基于偏好向量的距离支配(reference vector based distance dominance,rd-dominance)规则,解决了传统r支配(reference solution based dominance,r-dominance)受偏好点位置影响的问题。同时,设计了一种自适应的算法阈值更新机制,使得算法在进化前期可以充分搜索靠近Pareto前沿面的个体,保证了种群的多样性。然后,设计了一种基于偏好角度的偏好区域划分方法并将其与所提rd支配规则融合,提出了一种基于角度和距离混合引导策略。在进化中后期,利用所提偏好区域划分方法对偏好区域进行划分,仅对偏好区域内的个体进行rd支配排序,从而快速引导种群向着决策者感兴趣的区域进化,提高了算法的优化效率。在标准测试函数上的实验结果表明,与几种典型的偏好多目标进化算法相比,所提算法给出的优化结果具有更好的收敛性和稳定性,且不受偏好点位置的影响。同时,所提算法与传统基于r支配的算法相比具有更快的收敛速度。Considering the problems of algorithm performance being affected by the locations of preference points,having difficulty in controlling the size of the preference solution set,and slow convergence speed of traditional preference multi-objective evolutionary algorithms,a preferred multi-objective evolutionary algorithm based on a hybrid angle and distance guidance strategy is proposed.Firstly,a reference vector based distance dominance(rd-dominance)rule based on preference vector is designed to solve the problem that traditional reference solution based dominance(r-dominance)is affected by the position of the preference point.At the same time,an adaptive algorithm threshold update mechanism is designed so that the algorithm can fully search for individuals close to the Pareto front in the early stage of evolution,ensuring the diversity of the population.Then,a preference area division method based on preference angle is designed and integrated with the proposed rd-dominance rule,and a hybrid guidance strategy based on angle and distance is proposed.In the middle and late stages of evolution,the proposed preference area division method is used to divide the preference area,and only the individuals in the preference area are ranked by rd-dominance,thereby quickly guiding the population to evolve toward the area that the decision maker is interested in,and improving the optimization efficiency of the algorithm.Experimental results on standard test functions show that compared with several typical preference multi-objective evolutionary algorithms,the optimization results given by the proposed algorithm have better convergence and stability and are not affected by the position of the preference point.At the same time,the proposed algorithm has faster convergence speed than the traditional algorithm based on r-dominance.

关 键 词：多目标优化 混合引导 偏好信息 支配排序 

分 类 号：TP180[自动化与计算机技术—控制理论与控制工程]