Scholarlink

<p>Precise forecasting of debris flow volume is critical for mitigating its catastrophic impacts in mountainous areas. Traditional machine learning models often exhibit low accuracy and instability, especially with small datasets. Likewise, current meta-heuristic optimization algorithms need improved balance between exploration and exploitation. To address these limitations, this study proposes a hybrid model (IWAA-BPNN) combining an improved weighted average algorithm (IWAA) and back propagation neural network (BPNN) for high-precision debris flow volume prediction. Dimensionality reduction was performed using Spearman's correlation coefficient and maximal information coefficient, identifying five key factors. The WAA algorithm was improved through good point set initialization, adaptive weight adjustment, and spiral search strategies, which optimized BPNN parameters and enhanced both global and local search capabilities. Validation with real-world debris flow cases from Beichuan County showed the IWAA-BPNN model achieved exceptional performance (<i>R</i><sup>2</sup> = 98.5%). Comparative analysis with mainstream BPNN-integrated algorithms (PSO-BPNN and WOA-BPNN) and standalone BPNN showed <i>R</i><sup>2</sup> improvements of 6.62, 8.55, and 10.22%, respectively. Stability was assessed via five independent inner-loop validations, with results indicating the mean absolute percentage error remained below 7% across all tests. This highlights the robustness of the proposed model, offering a novel strategy for debris flow risk management in data-scarce scenarios. This study provides valuable insights for debris flow prevention in small-sample scenarios and contributes to machine learning algorithm development.</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

Debris flow volume prediction model based on a hybrid BP neural network optimized by adaptive weighted average algorithm with spiral search enhancement

  • 期刊论文

  • 发表日期：

    2026年7月8日

查看全文
  • Xi Wang,
  • Li Li,
  • Yue Qiang,
  • Chen Cheng,
  • Xinyi Tan,
  • Nanxi Chen,
  • Xing Wang

摘要

Precise forecasting of debris flow volume is critical for mitigating its catastrophic impacts in mountainous areas. Traditional machine learning models often exhibit low accuracy and instability, especially with small datasets. Likewise, current meta-heuristic optimization algorithms need improved balance between exploration and exploitation. To address these limitations, this study proposes a hybrid model (IWAA-BPNN) combining an improved weighted average algorithm (IWAA) and back propagation neural network (BPNN) for high-precision debris flow volume prediction. Dimensionality reduction was performed using Spearman's correlation coefficient and maximal information coefficient, identifying five key factors. The WAA algorithm was improved through good point set initialization, adaptive weight adjustment, and spiral search strategies, which optimized BPNN parameters and enhanced both global and local search capabilities. Validation with real-world debris flow cases from Beichuan County showed the IWAA-BPNN model achieved exceptional performance (R2 = 98.5%). Comparative analysis with mainstream BPNN-integrated algorithms (PSO-BPNN and WOA-BPNN) and standalone BPNN showed R2 improvements of 6.62, 8.55, and 10.22%, respectively. Stability was assessed via five independent inner-loop validations, with results indicating the mean absolute percentage error remained below 7% across all tests. This highlights the robustness of the proposed model, offering a novel strategy for debris flow risk management in data-scarce scenarios. This study provides valuable insights for debris flow prevention in small-sample scenarios and contributes to machine learning algorithm development.