文献类型：期刊文献

英文题名：A Dual-Population Constrained Multi-Objective Evolutionary Algorithm with Adaptive Knowledge Migration

作者：Yang, Youliang[1];Xu, Sijia[2];Xu, Yang[2];Shi, Wanxin[2];Yang, He[3];Ding, Weichao[2]

机构：[1]China Petr & Chem Corp, Beijing 100728, Peoples R China;[2]East China Univ Sci & Technol, Sch Informat Sci & Engn, Shanghai 200237, Peoples R China;[3]Sinopec Res Inst Petr Proc Co Ltd, Beijing 100083, Peoples R China

年份：2026

卷号：14

期号：12

外文期刊名：PROCESSES

收录：;EI(收录号：20262621013693);WOS:【SCI-EXPANDED(收录号:WOS:001802347400001)】；

基金：This research was funded by the National Natural Science Foundation of China (Grant No. 62402118 and 62403201), the Shanghai Magnolia Talent Program Pujiang Project (Grant No. 24PJD003), the "Lei Tai" Action Plan (Guangxi Science and Technology Innovation Platform Program)-Guangxi Laboratory Capacity Building (Contract No. LT2504240026), and the Shanghai Pilot Program for Basic Research (Grant No. 22TQ1400100-16).

语种：英文

外文关键词：constrained multi-objective optimization; dual-population coevolution; adaptive knowledge migration; negative-transfer suppression; evolutionary algorithm

摘要：Constrained multi-objective optimization problems (CMOPs) widely exist in scientific research and industrial applications. In Type IV CMOPs, where the constrained Pareto front (CPF) is significantly separated from the unconstrained Pareto front (UPF) by large infeasible barriers, traditional single-population evolutionary algorithms often suffer from severe search reachability difficulties. Moreover, while existing dual-population coevolutionary frameworks can exploit auxiliary populations to provide global guidance for obstacle crossing, they typically adopt a constant knowledge transfer intensity, which may introduce negative transfer and interfere with fine-grained CPF convergence in later evolutionary stages. To address these challenges, this paper proposes a Dual-Population Constrained Multi-Objective Evolutionary Algorithm with Adaptive Knowledge Migration (ADCMO). The algorithm constructs a main-auxiliary dual-population coevolutionary framework: the main population pursues feasible convergence under the original constraints, while the auxiliary population explores the unconstrained objective landscape to maintain global awareness. A linearly decaying migration control factor is introduced to dynamically regulate the intensity of cross-population knowledge transfer. Specifically, a dual-defense mechanism is established by simultaneously controlling the auxiliary participation ratio in mating pool construction and the auxiliary offspring injection scale in environmental selection, thereby achieving the synergistic effect of enhanced obstacle crossing in early evolution and progressive interference suppression in later stages. Extensive experiments on two benchmark suites comprising 23 test problems and ten representative real-world constrained multi-objective optimization problems demonstrate that ADCMO shows clear advantages on several large-barrier Type IV-like CMOPs, especially on the LIR-CMOP suite, while maintaining feasible and competitive behavior on most remaining instances. Ablation studies further verify the non-negligible contributions of the auxiliary population, the adaptive migration factor, and the dual-defense mechanism to the overall performance.