Scholarlink

<p>To enhance global search capability, and convergence efficiency of the black kite algorithm (BKA), this paper proposes a multi-strategy improved black kite algorithm (SWBKA). First, latin hypercube sampling is introduced to improve the initial population quality. Second, a fused Cauchy-Gaussian disturbance and adaptive spiral position update strategy is proposed to balance the global search and local exploitation capabilities during the migration behavior. Finally, a lens imaging reverse learning elite strategy is incorporated to help the algorithm effectively avoid local optima. Ablation experiments on the three strategies were conducted using 13 benchmark functions, and the performance of the improved algorithm was validated on 42 functions. Experimental results show that the proposed SWBKA achieves an overall average improvement of 47.38% in optimal values compared to BKA, demonstrating significant superiority. Furthermore, the proposed SWBKA was used to optimize the hyperparameters of the gated recurrent unit neural network, followed by prediction experiments on an actual electric load dataset. The experimental results demonstrate the algorithm’s capability to solve real-world problems. The code has been released in <a href="https://github.com/Daniel-Z-H-Dnn/SWBKA">https://github.com/Daniel-Z-H-Dnn/SWBKA</a>.</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

Multi-strategy improved black-winged kite algorithm and its application in short-term load forecasting using gated recurrent unit neural networks

  • 期刊论文

  • 发表日期：

    2025年6月25日

查看全文
  • Hao Zhang,
  • Liming Liu,
  • Yongkuan Yang,
  • Weiming Shao

摘要

To enhance global search capability, and convergence efficiency of the black kite algorithm (BKA), this paper proposes a multi-strategy improved black kite algorithm (SWBKA). First, latin hypercube sampling is introduced to improve the initial population quality. Second, a fused Cauchy-Gaussian disturbance and adaptive spiral position update strategy is proposed to balance the global search and local exploitation capabilities during the migration behavior. Finally, a lens imaging reverse learning elite strategy is incorporated to help the algorithm effectively avoid local optima. Ablation experiments on the three strategies were conducted using 13 benchmark functions, and the performance of the improved algorithm was validated on 42 functions. Experimental results show that the proposed SWBKA achieves an overall average improvement of 47.38% in optimal values compared to BKA, demonstrating significant superiority. Furthermore, the proposed SWBKA was used to optimize the hyperparameters of the gated recurrent unit neural network, followed by prediction experiments on an actual electric load dataset. The experimental results demonstrate the algorithm’s capability to solve real-world problems. The code has been released in https://github.com/Daniel-Z-H-Dnn/SWBKA.