Multi Actors-Critic based particle swarm optimization algorithm

作     者：Chen, Hui Shen, Li-Yong Wang, Chunmei Tian, Lei Zhang, Sanguo 

作者机构：Univ Chinese Acad Sci Sch Math Sci Beijing 100049 Peoples R China Jinhang Digital Technol Co Ltd Beijing 100028 Peoples R China 

出 版 物：《NEUROCOMPUTING》 (Neurocomputing)

年 卷 期：2025年第624卷

核心收录：

学科分类：08[工学] 0812[工学-计算机科学与技术（可授工学、理学学位）] 

基　　金：Fundamental Research Funds for the Central Universities Chinese Academy of Sciences, CAS, (XDB0640200) National Natural Science Foundation of China, NSFC, (12371384) 

主　　题：Particle swarm optimization Parameter control Reinforcement learning Multi-agent 

摘      要：This paper introduces the Multi-Actor-Critic-based Particle Swarm Optimization Algorithm (MACPSO), an innovative approach to optimizing particle performance through dynamic parameter control. MACPSO applies a multi-actor-critic framework to enhance exploration and maintain population diversity. It features multiple actor networks for adaptive parameter adjustment, coupled with a single critic network to guide the shared optimization goal. The algorithm incorporates particle grouping for intra-group updates and an inter-group updating scheme to facilitate the exchange of optimal information. Furthermore, MACPSO integrates a mutation mechanism aimed at improving the performance of the least effective particles, thereby ensuring sustained diversity within the population. To validate the superior performance of the proposed algorithm, a comparative study was conducted with nine advanced PSO variants using the CEC2017 benchmark suite. The experimental results demonstrate that MACPSO achieved the highest overall ranking in terms of average error across twenty-nine different benchmark functions. Additionally, numerical, graphical, and statistical analyses confirm that MACPSO outperforms the other nine PSO variants in terms of both effectiveness and robustness.