基于Stacking-SHAP的煤自燃倾向性影响因素研究  

Research on Influencing Factors of Coal Spontaneous Combustion Tendency Based on Stacking-SHAP

作　　者：崔忠麒 徐娅煊 苏皓 CUI Zhongqi;XU Yaxuan;SU Hao(North China University of Science and Technology,Tangshan 063210,China)

机构地区：[1]华北理工大学,河北唐山063210

出　　处：《煤炭技术》2025年第1期150-155,共6页Coal Technology

基　　金：唐山市基础研究科技计划项目(22130209H);国家级大学生创新创业训练计划项目(202310081005)。

摘　　要：为对煤自燃倾向性做出准确的预测,挖掘不同煤样属性对煤自燃倾向性的贡献程度,提出基于Stacking-SHAP的煤自燃倾向性预测模型。分别将煤体自身属性及其自燃倾向性综合判定指数作为模型输入和输出。该模型融合支持向量回归(SVR)、极限梯度提升归回树(XGBoost)、随机森林(RandomForest)、梯度提升决策树(GBDT),并利用网格搜索法对各基础模型参数进行优化,同时结合SHAP算法对不同影响因素的贡献度进行计算。结果显示,优化后的SVR、XGBoost、RF、GBDT和Stacking的判定系数R^(2)分别为0.933、0.887、0.950、0.925、0.984。在煤自燃倾向性影响因素中,重要性程度靠前的特征依次是氧含量、挥发分含量、脂肪烃峰面积值、C/H、羟基峰面积值以及总孔体积共6种特征。模型的建立为煤自燃倾向性预测与煤自燃灾害防治提供了一种新方法。To accurately predict the spontaneous combustion tendency of coal,explore the contribution degree of different coal sample properties to coal spontaneous combustion tendeney,and propose a coal spontaneous combustion tendency prediction model based on Stacking-SHAP.Both the intrinsic properties of coal and the comprehensive judgment index of its spontaneous combustion propensity are taken as model inputs and outputs.The model integrates Support Vector Regression(SVR),Extreme Gradient Boosting(XGBoost),RandomForest(RF),and Gradient Boosting Decision Tree(GBDT),optimizing the parameters of each base model using grid search.Additionally,the SHAPalgorithm is employed to calculate the contribution of different influencing factors.The results show that the optimized R^(2)coefficients for SVR,XGBoost,RF,GBDT,and Stacking are0.933,0.887,0.950,0.925,and 0.984,respectively.Among the influencing factors of coal spontaneous combustion propensity,the most important features are oxygen content,volatile matter content,peak area of aliphatic hydrocarbons,C/H,peak area of hydroxyl groups,and total pore volume,totaling six features.The establishment of the model provides a new method for predicting coal spontaneous combustion propensity and preventing coal spontaneous combustion disasters.

关 键 词：煤自燃倾向性 STACKING SHAP 机器学习 数据挖掘 

分 类 号：TD75[矿业工程—矿井通风与安全]