Scholarlink

<p>The optimizer is essential for model training, accelerating training, improving stability, and finding globally optimal solutions. The Aquila optimizer (AO) is an intelligent algorithm, simulating the hunting behavior of North American eagles, and excels in finding optimal solutions. However, the classical AO suffers from low convergence accuracy and tends to fall into local optima in complex tasks. This paper introduces a novel and improved Aquila optimizer (I-AO) for population-based meta-inspired domains, enhancing global optimization reliability. We first introduce spatial configuration differences in distance and angle to replace the traditional iteration-based criteria. This approach enhances the optimizer's convergence speed and robustness, enabling quick adaptation to dynamic changes. Next, we introduce the improved random walk update (I-RWU), improved stochastic correction update (I-SCU), and improved dynamic weight update (I-DWU) strategies. These strategies enhance search diversity and balance exploration with exploitation, avoiding local optima. Thus, I-AO achieves higher computational efficiency and greater exploration potential for optimal convergence. Furthermore, performance is evaluated on the CEC2017 and CEC2019 benchmark functions, where I-AO improves objective function values by 57.2% and 54.5%, respectively, compared to the traditional AO. Additionally, the I-AO's generalizability is demonstrated in an engineering problem predicting the Remaining Useful Life (RUL) of air turbine starter (ATS) bearings, with improvements in MAE and RMSE of 59% and 73.1%. The superior capability of the proposed I-AO is demonstrated through corresponding experiments.</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

Improved Aquila optimizer and its applications

  • 期刊论文

  • 发表日期：

    2025年2月25日

查看全文
  • Runxia Guo,
  • Jingxu Yi

摘要

The optimizer is essential for model training, accelerating training, improving stability, and finding globally optimal solutions. The Aquila optimizer (AO) is an intelligent algorithm, simulating the hunting behavior of North American eagles, and excels in finding optimal solutions. However, the classical AO suffers from low convergence accuracy and tends to fall into local optima in complex tasks. This paper introduces a novel and improved Aquila optimizer (I-AO) for population-based meta-inspired domains, enhancing global optimization reliability. We first introduce spatial configuration differences in distance and angle to replace the traditional iteration-based criteria. This approach enhances the optimizer's convergence speed and robustness, enabling quick adaptation to dynamic changes. Next, we introduce the improved random walk update (I-RWU), improved stochastic correction update (I-SCU), and improved dynamic weight update (I-DWU) strategies. These strategies enhance search diversity and balance exploration with exploitation, avoiding local optima. Thus, I-AO achieves higher computational efficiency and greater exploration potential for optimal convergence. Furthermore, performance is evaluated on the CEC2017 and CEC2019 benchmark functions, where I-AO improves objective function values by 57.2% and 54.5%, respectively, compared to the traditional AO. Additionally, the I-AO's generalizability is demonstrated in an engineering problem predicting the Remaining Useful Life (RUL) of air turbine starter (ATS) bearings, with improvements in MAE and RMSE of 59% and 73.1%. The superior capability of the proposed I-AO is demonstrated through corresponding experiments.