Scholarlink

<p>Accurate prediction of coal spontaneous combustion temperature is the key to mine fire prevention and extinguishing. A novel method combining the Boruta feature selection algorithm with northern goshawk optimization (NGO) and gradient boosting decision tree (GBDT) was proposed. Based on programmed temperature experiments within 30–300&#xa0;°C, a multivariate indicator system comprising 8 single gas indicators and 29 gas ratios was established. The Boruta algorithm was applied to identify key features from this extensive set of indicators, while NGO was used to optimize model hyperparameters. A systematic comparison of four ensemble tree models of random forest, light gradient boosting machine, extremely randomized trees, and GBDT was made before and after optimization. The results showed that the Boruta–NGO–GBDT model achieved excellent performance with MSE (mean squared error) of 243.91 and <i>R</i><sup>2</sup> of 0.96. Compared with the unoptimized model, NGO reduced the MSE of GBDT by 89.07%. Shapley additive explanations analysis revealed O<sub>2</sub>/CO<sub>2</sub>, O<sub>2</sub>/CH<sub>4</sub>, CO/C<sub>3</sub>H<sub>8</sub>, C<sub>2</sub>H<sub>4</sub>/C<sub>3</sub>H<sub>8</sub>, and O<sub>2</sub>/CO as the most critical indicators, collectively accounting for over 50% of prediction importance. The model’s generalization capability was validated using coal samples from different mines, achieving a <i>R</i><sup>2</sup> of 0.949 on the testing set. The results indicated that the combination of gas indicators selected by the Boruta–NGO–GBDT model can accurately predict coal spontaneous combustion temperature. This offers practical guidance for monitoring and controlling coal oxidation processes, providing an effective early warning solution for coal mine safety.</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

Prediction of Coal Spontaneous Combustion Temperature Using Boruta–NGO–GBDT: A Machine Learning Approach to Feature Selection and Metaheuristic Optimization

  • 期刊论文

  • 发表日期：

    2026年3月25日

查看全文
  • Yan Zhang,
  • Hemeng Zhang,
  • Xiaoying Liu,
  • Shuliang Jia,
  • Yongjun Wang,
  • Hung Vo Thanh

摘要

Accurate prediction of coal spontaneous combustion temperature is the key to mine fire prevention and extinguishing. A novel method combining the Boruta feature selection algorithm with northern goshawk optimization (NGO) and gradient boosting decision tree (GBDT) was proposed. Based on programmed temperature experiments within 30–300 °C, a multivariate indicator system comprising 8 single gas indicators and 29 gas ratios was established. The Boruta algorithm was applied to identify key features from this extensive set of indicators, while NGO was used to optimize model hyperparameters. A systematic comparison of four ensemble tree models of random forest, light gradient boosting machine, extremely randomized trees, and GBDT was made before and after optimization. The results showed that the Boruta–NGO–GBDT model achieved excellent performance with MSE (mean squared error) of 243.91 and R2 of 0.96. Compared with the unoptimized model, NGO reduced the MSE of GBDT by 89.07%. Shapley additive explanations analysis revealed O2/CO2, O2/CH4, CO/C3H8, C2H4/C3H8, and O2/CO as the most critical indicators, collectively accounting for over 50% of prediction importance. The model’s generalization capability was validated using coal samples from different mines, achieving a R2 of 0.949 on the testing set. The results indicated that the combination of gas indicators selected by the Boruta–NGO–GBDT model can accurately predict coal spontaneous combustion temperature. This offers practical guidance for monitoring and controlling coal oxidation processes, providing an effective early warning solution for coal mine safety.