摘要
煤矿地表沉陷严重威胁矿区生态环境及周边基础设施安全,因此精准预计地表沉陷意义重大。但地表沉陷的预计复杂,概率积分法预计地表沉陷准确性较低。提出了一种基于改进的径向移动(Improved Radial Movement Optimization,IRMO)算法优化极致梯度提升(eXtreme Gradient Boosting,XGBoost)算法的地表沉陷预计模型,通过IRMO算法选择XGBoost算法中的学习率、正则化等超参数的最优值,提高了地表沉陷预计精度,并与遗传算法(Genetic Algorithm,GA)优化的XGBoost算法、XGBoost算法的预测结果进行了对比分析,IRMO-XGBoost模型的均方根误差R_(MSE)(0.156)和平均绝对误差M_(AE)(0.126)更低,决定系数R^(2)(0.970)更高。运用IRMO-XGBoost模型对建北煤矿4^(-2)煤305工作面的地表沉陷值进行了预测,结果表明,IRMO-XGBoost模型预测精度明显优于XGBoost算法。最后用Shapley解释(SHapley Additive exPlanations,SHAP)方法量化模型的输入特征对地表沉陷预测的贡献。基于IRMO-XGBoost构建的地表沉陷预计模型精度高,可以极大地帮助矿区掌握地表沉陷对地表环境的破坏程度,为矿区生态环境的保护管理和安全生产措施的制定提供超前预测。
Surface subsidence in coal mining areas poses significant risks to both the biological environment and the safety of nearby infrastructure.As a result,accurately predicting surface subsidence is vital for maintaining mining safety and promoting environmental sustainability.Traditional methods,such as the probability integral method,face challenges due to their inherent accuracy limitations and computational complexity in real-world applications.This study introduces a novel surface subsidence prediction model that leverages the Improved Radial Movement Algorithm(IRMO)to enhance the performance of the eXtreme Gradient Boosting(XGBoost)algorithm.This model utilizes geological mining conditions as input features and surface subsidence values as output features.The IRMO is employed to identify the optimal hyperparameters,such as learning rate and regularization,for the XGBoost algorithm,significantly enhancing the accuracy of surface subsidence predictions.Specifically,four key XGBoost hyperparameters are globally optimized using the IRMO algorithm:gamma value(range:[0,10]),learning rate(range:[0.01,0.5]),minimum child weight(range:[1,20]),and maximum tree depth(range:[1,10]).The performance of the IRMO-XGBoost prediction model was compared with both the standard XGBoost and the GA-XGBoost(XGBoost optimized by the genetic algorithm).The experimental results demonstrate that IRMO-XGBoost delivers exceptional performance,with the following indicators:R_(MSE)=0.156,M_(AE)=0.126,and R^(2)=0.970.When compared to GA-XGBoost,IRMO-XGBoost reduced R_(MSE) by 13.3%(R_(MSE):0.180)and M_(AE) by 11.3%(M_(AE):0.142).Furthermore,when compared to the standard XGBoost model,IRMO-XGBoost achieved a 59.2%reduction in R_(MSE)(R_(MSE):0.382)and a 62.3%reduction in M_(AE)(M_(AE):0.334).The IRMO-XGBoost model was employed to predict the surface subsidence value of working face 305 in the Coal 4^(-2) section of the Jianbei Coal Mine.The results indicate that the prediction accuracy of the IRMO-XGBoost model significantly outperformed that of the standard XGBoost algorithm.Additionally,the contribution of input features to the surface subsidence predictions was analyzed using the Shapley Additive exPlanations(SHAP)interpretation method.With its ability to provide interpretable feature contributions alongside high-precision and computationally efficient subsidence predictions,this study offers valuable technical support for assessing environmental risks associated with mining and developing effective mitigation strategies.
作者
王军胜
王宏涛
张文
白宇
金亮星
高志勇
刘娉婷
WANG Junsheng;WANG Hongtao;ZHANG Wen;BAI Yu;JIN Liangxing;GAO Zhiyong;LIU Pingting(Shaanxi Coal Industry Group Huangling Jianzhuang Mining Co.,Ltd.,Yan'an 727300,Shaanxi,China;School of Civil Engineering,Central South University,Changsha 410075,China)
出处
《安全与环境学报》
北大核心
2025年第9期3504-3513,共10页
Journal of Safety and Environment
关键词
安全工程
地表沉陷预计
改进的径向移动算法
极致梯度提升算法
概率积分法
safety engineering
surface subsidence prediction
improved radial movement optimization algorithm
extreme gradient boosting algorithm
probability integral method
作者简介
王军胜,高级工程师,从事煤矿地质、安全采矿研究,1783077505@qq.com。
