Scholarlink

<p>This paper introduces a Reinforcement Learning-based Shuffled Multi Opposition-Based Learning Evolutionary Algorithm (RLSMOBEA) and its enhanced variant, L-RLSMOBEA, which incorporates Linear Population Size Reduction (LPSR). Shuffled evolutionary algorithms partition a population into groups, each undergoing an independent evolutionary process. The proposed framework addresses key limitations in shuffled evolutionary algorithms, such as premature convergence and inefficient operator selection, by synergistically integrating three core mechanisms. First, Opposition-Based Learning (OBL) is embedded into the evolutionary processes of Shuffled Frog Leaping (SFL), Shuffled Complex Evolution (SCE), and Shuffled Differential Evolution (SDE) to enhance exploration and accelerate convergence. Second, a Q-learning agent is employed as a hyper-heuristic to intelligently select the most promising evolutionary strategy at each iteration. Third, a Linear Population Size Reduction (LPSR) strategy dynamically reduces the population and number of groups to efficiently manages computational resources. The performance of L-RLSMOBEA is rigorously evaluated on the CEC2014 and CEC2017 benchmark suites. Evaluation results, supported by statistical analysis, demonstrate that it significantly outperforms its predecessors (SMEA, SFL, SCE, SDE) and shows competitive or superior performance against state-of-the-art algorithms. Statistical significance tests, computational complexity analysis and a comprehensive ablation study confirm the individual and combined efficacy of the RL, OBL, and LPSR components.</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

A reinforcement learning hyper-heuristic for opposition-enhanced shuffled multi-strategy evolutionary algorithms with adaptive population sizing

  • 期刊论文

  • 发表日期：

    2026年6月2日

查看全文
  • Morteza Alinia Ahandani,
  • Hosein Alavi-Rad,
  • Hamed Kharrti,
  • Mehrdad Saif,
  • Mohammad Reza Chalak Qazani

摘要

This paper introduces a Reinforcement Learning-based Shuffled Multi Opposition-Based Learning Evolutionary Algorithm (RLSMOBEA) and its enhanced variant, L-RLSMOBEA, which incorporates Linear Population Size Reduction (LPSR). Shuffled evolutionary algorithms partition a population into groups, each undergoing an independent evolutionary process. The proposed framework addresses key limitations in shuffled evolutionary algorithms, such as premature convergence and inefficient operator selection, by synergistically integrating three core mechanisms. First, Opposition-Based Learning (OBL) is embedded into the evolutionary processes of Shuffled Frog Leaping (SFL), Shuffled Complex Evolution (SCE), and Shuffled Differential Evolution (SDE) to enhance exploration and accelerate convergence. Second, a Q-learning agent is employed as a hyper-heuristic to intelligently select the most promising evolutionary strategy at each iteration. Third, a Linear Population Size Reduction (LPSR) strategy dynamically reduces the population and number of groups to efficiently manages computational resources. The performance of L-RLSMOBEA is rigorously evaluated on the CEC2014 and CEC2017 benchmark suites. Evaluation results, supported by statistical analysis, demonstrate that it significantly outperforms its predecessors (SMEA, SFL, SCE, SDE) and shows competitive or superior performance against state-of-the-art algorithms. Statistical significance tests, computational complexity analysis and a comprehensive ablation study confirm the individual and combined efficacy of the RL, OBL, and LPSR components.