Scholarlink

<p>Metaheuristic optimization algorithms have been broadly implemented to complicated numerical optimization problems; nevertheless, numerous current approaches hurt from limited adaptability and an unevenness between exploration and exploitation, particularly on complex benchmark landscapes. To address these problems, this work focuses on enhancing the newly proposed American Zebra-optimization algorithm Method (AZOA), which is modelled on the zebra herd behaviour. Despite the fact that AZOA exhibits strong exploration potential, it faces specific drawbacks, including insufficient exploitation and a likelihood to become confined in local optima when overcoming intricate optimization tasks. To address these tasks, this works aims to improve the AZOA by incorporating a Q-Learning–based adaptive decision mechanism, resulting in a learning-augmented variant referred to as QLAZOA. This novel mechanism allows each zebra to learn from previous actions and select suitable movement strategies based on reward feedback, effectively improving the stability between exploitation and exploration. The suggested QLAZOA is tested using classical benchmark functions (CEC2005), recent test suites (CEC2019), and real-world engineering design case studies. Additionally, statistical validation, including the t-test, is done to examine the importance and dependability of the outcomes. Experimental results show that QLAZOA achieves strong and competitive rankings on the CEC2005 benchmark suite, including both the full set of 23 functions and a subset of 13 functions involving top-performing optimizers, where it outperforms several classical and swarm-based algorithms. On the more challenging CEC2019 benchmark suite, QLAZOA exhibits benchmark-dependent performance, remaining competitive with multiple swarm-based optimizers. Overall, the results indicate that QLAZOA provides a learning-augmented optimization framework with clear advantages on classical and moderately complex landscapes. These findings offer insight into the role of reinforcement learning in metaheuristic optimization and suggests various engineering applications.</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

Q-Learning American Zebra Optimization Algorithm for Global Optimization

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年5月18日

查看全文
  • Rakesh Kumar Olipalli,
  • Sriram Naraharisetti,
  • Rohith Perumalla,
  • Jathin Vallabhapurapu,
  • Prudhvi Naidu Thota,
  • Sarada Mohapatra,
  • Prabhujit Mohapatra

摘要

Metaheuristic optimization algorithms have been broadly implemented to complicated numerical optimization problems; nevertheless, numerous current approaches hurt from limited adaptability and an unevenness between exploration and exploitation, particularly on complex benchmark landscapes. To address these problems, this work focuses on enhancing the newly proposed American Zebra-optimization algorithm Method (AZOA), which is modelled on the zebra herd behaviour. Despite the fact that AZOA exhibits strong exploration potential, it faces specific drawbacks, including insufficient exploitation and a likelihood to become confined in local optima when overcoming intricate optimization tasks. To address these tasks, this works aims to improve the AZOA by incorporating a Q-Learning–based adaptive decision mechanism, resulting in a learning-augmented variant referred to as QLAZOA. This novel mechanism allows each zebra to learn from previous actions and select suitable movement strategies based on reward feedback, effectively improving the stability between exploitation and exploration. The suggested QLAZOA is tested using classical benchmark functions (CEC2005), recent test suites (CEC2019), and real-world engineering design case studies. Additionally, statistical validation, including the t-test, is done to examine the importance and dependability of the outcomes. Experimental results show that QLAZOA achieves strong and competitive rankings on the CEC2005 benchmark suite, including both the full set of 23 functions and a subset of 13 functions involving top-performing optimizers, where it outperforms several classical and swarm-based algorithms. On the more challenging CEC2019 benchmark suite, QLAZOA exhibits benchmark-dependent performance, remaining competitive with multiple swarm-based optimizers. Overall, the results indicate that QLAZOA provides a learning-augmented optimization framework with clear advantages on classical and moderately complex landscapes. These findings offer insight into the role of reinforcement learning in metaheuristic optimization and suggests various engineering applications.