Scholarlink

<p>Inspired by natural migration phenomena, this study proposes a new algorithm named the Migration Model based on PSO (MMPSO). The algorithm operates within a multi-swarm structure to emulate an ecological community. It allows sub-swarms to autonomously determine their movement strategies while embedding a natural selection mechanism. Consequently, MMPSO can maintain and diversify multiple search strategies simultaneously. The paper proposes and integrates three new techniques as modular components, thereby forming the MMPSO algorithm: (i) Multi-Vertex Convergent Search (MVCS), which initializes the population clustered near the vertices of the hyper-rectangle in the solution space, thereby enlarging the initial exploration region and improving the ability to handle multimodal problems; (ii) Selective Crossover for Superior Traits (SCST), which mimics natural selection by identifying underperforming swarms, selecting weak individuals, and regenerate them using superior traits, extracted from the history archive - the analysis further indicates that SCST is a key component that markedly improves swarm adaptability and convergence quality.; and (iii) Auto-Transition Search Mode works as a “control valve” that adaptively switches each swarm between an Exploration mode (global diversification) and an Exploitation mode (local refinement) based on its convergence status, so that different swarms can operate under these two strategies in parallel. The guiding principles of MMPSO are diversity, adaptability, and the strategic independence of each subswarm. In the exploitation mode, MMPSO extends the update mechanism of classical PSO, while introducing time-dependent parameters. On the other hand, in the exploration mode, every subswarm has the ability to choose among four update scenarios, each characterized by different combinations of random perturbation, swarm concentration, and Lévy flight–based movement. The experiments demonstrate that MMPSO is highly effective on SOP benchmarks, the 13 classical engineering problems, and the large-scale 942-bar truss tower. Furthermore, on the CEC2020 and CEC2017 benchmarks it maintains stable and competitive performance against comparative algorithms. The results validate that the new algorithm with the proposed mechanisms enhances solution quality across various problem classes.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

An adaptive multi-swarm optimization algorithm inspired by migration for solving multimodal problems

  • 期刊论文

  • 发表日期：

    2026年2月4日

查看全文
  • Tri Ton-That,
  • Minh Hoang-Le,
  • Cuong Ngo-Huu,
  • Thanh Cuong-Le

摘要

Inspired by natural migration phenomena, this study proposes a new algorithm named the Migration Model based on PSO (MMPSO). The algorithm operates within a multi-swarm structure to emulate an ecological community. It allows sub-swarms to autonomously determine their movement strategies while embedding a natural selection mechanism. Consequently, MMPSO can maintain and diversify multiple search strategies simultaneously. The paper proposes and integrates three new techniques as modular components, thereby forming the MMPSO algorithm: (i) Multi-Vertex Convergent Search (MVCS), which initializes the population clustered near the vertices of the hyper-rectangle in the solution space, thereby enlarging the initial exploration region and improving the ability to handle multimodal problems; (ii) Selective Crossover for Superior Traits (SCST), which mimics natural selection by identifying underperforming swarms, selecting weak individuals, and regenerate them using superior traits, extracted from the history archive - the analysis further indicates that SCST is a key component that markedly improves swarm adaptability and convergence quality.; and (iii) Auto-Transition Search Mode works as a “control valve” that adaptively switches each swarm between an Exploration mode (global diversification) and an Exploitation mode (local refinement) based on its convergence status, so that different swarms can operate under these two strategies in parallel. The guiding principles of MMPSO are diversity, adaptability, and the strategic independence of each subswarm. In the exploitation mode, MMPSO extends the update mechanism of classical PSO, while introducing time-dependent parameters. On the other hand, in the exploration mode, every subswarm has the ability to choose among four update scenarios, each characterized by different combinations of random perturbation, swarm concentration, and Lévy flight–based movement. The experiments demonstrate that MMPSO is highly effective on SOP benchmarks, the 13 classical engineering problems, and the large-scale 942-bar truss tower. Furthermore, on the CEC2020 and CEC2017 benchmarks it maintains stable and competitive performance against comparative algorithms. The results validate that the new algorithm with the proposed mechanisms enhances solution quality across various problem classes.