Institute of Resources & Environment, Henan Polytechnic University, Jiaozuo, 454000, China.
Collaborative Innovation Center of Coal Work Safety and Clean High Efficiency Utilization, Jiaozuo, 454000, China.
Environ Monit Assess. 2024 Sep 2;196(10):876. doi: 10.1007/s10661-024-13040-z.
Mine water surge is one of the main safety risks in coal mines. This research offers a novel mine water source identification model (BO-CatBoost) to successfully avoid and control mine sudden water catastrophes by properly identifying the sources of mine water. First, the classification model is trained and built using the Categorical Boosting (CatBoost) algorithm. The Gaussian process Bayesian optimization (BO) algorithm is used to optimize parameters, and the optimal parameter combination is integrated into the CatBoost algorithm to build the BO-CatBoost mine water source identification model, which further improves the accuracy of mine water source identification. The model was also applied to the Pingdingshan mine to verify the practicality of the model. Then, 29 groups of unknown water sources in Pingdingshan were selected as validation samples for the model and compared with the conventional CatBoost, Light Gradient Boosting Machine (LightGBM), and Extreme Gradient Boosting (Xgboost) models. The comparison results demonstrate that the accuracy of LightGBM, Xgboost, CatBoost, and BO-CatBoost models can reach 69%, 79.3%, 79.3%, and 100% respectively, and the RMSE is 0.947, 0.643, 0.719, and 0.0 respectively. The comprehensive analysis shows that, when it comes to mine water source detection, the BO-CatBoost model performs noticeably better than other models in terms of discriminative accuracy and generalization capacity. Lastly, the multi-output prediction and decision-making process of the BO-CatBoost water source identification model is visualized by the interpretability analysis performed with the SHAP approach. The research demonstrates that the BO-CatBoost model can more precisely and impartially identify mine water sources, offering fresh concepts for mine water source detection.
矿井突水是煤矿的主要安全隐患之一。本研究提出了一种新颖的矿井水源识别模型（BO-CatBoost），通过正确识别矿井水源，可以成功避免和控制矿井突水灾害。首先，使用分类模型的 CatBoost 算法进行训练和构建。使用高斯过程贝叶斯优化（BO）算法优化参数，并将最优参数组合集成到 CatBoost 算法中，构建 BO-CatBoost 矿井水源识别模型，进一步提高了矿井水源识别的准确性。模型还应用于平顶山矿进行了验证，以验证模型的实用性。然后，选择平顶山的 29 组未知水源作为模型的验证样本，并与传统的 CatBoost、Light Gradient Boosting Machine（LightGBM）和 Extreme Gradient Boosting（Xgboost）模型进行比较。比较结果表明，LightGBM、Xgboost、CatBoost 和 BO-CatBoost 模型的准确率分别可以达到 69%、79.3%、79.3%和 100%，RMSE 分别为 0.947、0.643、0.719 和 0.0。综合分析表明，在矿井水源检测方面，BO-CatBoost 模型在判别精度和泛化能力方面明显优于其他模型。最后，使用 SHAP 方法进行可解释性分析，可视化了 BO-CatBoost 水源识别模型的多输出预测和决策过程。研究表明，BO-CatBoost 模型可以更准确、公正地识别矿井水源，为矿井水源检测提供了新的思路。
