Scholarlink

<p>Temperature and carbon monoxide (CO) concentration are key indicators in coal spontaneous combustion monitoring. To achieve accurate prediction of coal spontaneous combustion hazards underground, this study addressed the issue of insufficient accuracy in traditional coal spontaneous combustion prediction models by proposing a machine learning model that integrates multiple gas indicators with temperature features. A dataset comprised of features such as CO, temperature, O<sub>2</sub>, and olefins was constructed using coal spontaneous combustion temperature ramp-up experiments. Spearman's correlation analysis was employed to analyze the correlation among these features. The pelican optimization algorithm was improved (IPOA) by incorporating dynamic convergence factors, a memory pool, and hybrid search functionality to optimize long short-term memory (LSTM) hyperparameters. The results showed that the accuracy of the IPOA-LSTM model was 15–30% better than that of other mainstream models. SHAP (SHapley Additive exPlanations) analysis revealed that C<sub>2</sub>H<sub>4</sub>, O<sub>2</sub>, and CO/O<sub>2</sub> were key predictive variables, consistent with the results of the correlation analysis, validating the rationality of feature selection and the consistency of the model's internal mechanisms. The model demonstrated high accuracy and efficiency, providing reliable support for early warning of spontaneous coal combustion.</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 POA-LSTM Model Based on Multi-source Gas and Temperature Features for Predicting Key Indicators of Coal Spontaneous Combustion

  • 期刊论文

  • 发表日期：

    2026年1月3日

查看全文
  • Bo Tan,
  • Longkun Sui,
  • Liyang Gao,
  • Kuan Yang,
  • Songlu Tang,
  • Yunfei Zuo

摘要

Temperature and carbon monoxide (CO) concentration are key indicators in coal spontaneous combustion monitoring. To achieve accurate prediction of coal spontaneous combustion hazards underground, this study addressed the issue of insufficient accuracy in traditional coal spontaneous combustion prediction models by proposing a machine learning model that integrates multiple gas indicators with temperature features. A dataset comprised of features such as CO, temperature, O2, and olefins was constructed using coal spontaneous combustion temperature ramp-up experiments. Spearman's correlation analysis was employed to analyze the correlation among these features. The pelican optimization algorithm was improved (IPOA) by incorporating dynamic convergence factors, a memory pool, and hybrid search functionality to optimize long short-term memory (LSTM) hyperparameters. The results showed that the accuracy of the IPOA-LSTM model was 15–30% better than that of other mainstream models. SHAP (SHapley Additive exPlanations) analysis revealed that C2H4, O2, and CO/O2 were key predictive variables, consistent with the results of the correlation analysis, validating the rationality of feature selection and the consistency of the model's internal mechanisms. The model demonstrated high accuracy and efficiency, providing reliable support for early warning of spontaneous coal combustion.