Scholarlink

<p>Roof collapse in underground mining generate substantial economic while posing significant risks to worker safety. Accurate and interpretable roof pressure prediction in longwall mining faces is therefore paramount for mitigating risk. However, the non-stationary of field data, coupled with the “black box” nature of deep learning models, has long hindered predictive reliability and interpretability. To mitigate these challenges, this study proposes a hybrid modelling framework that combines the Discrete Wavelet Transform (DWT) to handling data non-stationarity with SHapley Additive Explanations (SHAP) to provide granular insights into feature contributions. Empirical evaluations across four deep learning architectures: Long Short-Term Memory (LSTM), Gated Recurrent Units (GRU), Transformer, and Autoformer, demonstrate the hybrid model’s performance compared to baseline methods. Additionally, SHAP-based analysis enhances transparency by identifying key parameters that drive model accuracy, thereby increasing domain relevance. Validation using datasets from Chinese coal mines substantiates the robustness of this approach, establishing it as a reliable and interpretable tool for roof pressure prediction in complex underground settings.</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

Hybrid discrete wavelet transform approach with explainable artificial intelligence for accurate prediction of non-stationary longwall roof pressure data

  • 期刊论文

  • 发表日期：

    2025年4月10日

查看全文
  • Dashu Yin,
  • Haoqian Chang,
  • Xiangqian Wang,
  • Huizong Li

摘要

Roof collapse in underground mining generate substantial economic while posing significant risks to worker safety. Accurate and interpretable roof pressure prediction in longwall mining faces is therefore paramount for mitigating risk. However, the non-stationary of field data, coupled with the “black box” nature of deep learning models, has long hindered predictive reliability and interpretability. To mitigate these challenges, this study proposes a hybrid modelling framework that combines the Discrete Wavelet Transform (DWT) to handling data non-stationarity with SHapley Additive Explanations (SHAP) to provide granular insights into feature contributions. Empirical evaluations across four deep learning architectures: Long Short-Term Memory (LSTM), Gated Recurrent Units (GRU), Transformer, and Autoformer, demonstrate the hybrid model’s performance compared to baseline methods. Additionally, SHAP-based analysis enhances transparency by identifying key parameters that drive model accuracy, thereby increasing domain relevance. Validation using datasets from Chinese coal mines substantiates the robustness of this approach, establishing it as a reliable and interpretable tool for roof pressure prediction in complex underground settings.