基于随机搜索优化XGBoost的瓦斯涌出量预测模型  

Prediction model of gas emission amount based on XGBoost optimized with random search algorithm

作　　者：马恒[1,2] 任美学 高科[1,2] MA Heng;REN Meixue;GAO Ke(College of Safety Science and Engineering,Liaoning Technical University,Huludao Liaoning 125105,China;Key Laboratory of Mine Thermodynamic Disaster and Control of Ministry of Education,Huludao Liaoning 125105,China)

机构地区：[1]辽宁工程技术大学安全科学与工程学院,辽宁葫芦岛125105 [2]矿山热动力灾害与防治教育部重点实验室,辽宁葫芦岛125105

出　　处：《中国安全生产科学技术》2022年第5期129-134,共6页Journal of Safety Science and Technology

基　　金：国家自然科学基金项目(52074148)。

摘　　要：为防治瓦斯灾害,解决井下瓦斯涌出量在预测过程中因影响因素繁多带来的精度较低问题,提出1种基于套索(Lasso)回归与随机搜索优化极限梯度提升(XGBoost)的模型进行瓦斯涌出量预测。以沈阳某煤矿综采面瓦斯涌出量历史数据为例,综合考虑影响瓦斯涌出量的影响因素。首先利用Lasso回归提取对瓦斯涌出量有重要影响的特征数据,作为预测输入;采用随机搜索算法对XGBoost模型4种主要参数进行寻优,选取最优参建立预测模型获得预测指标并分析比较其他模型。研究结果表明:Lasso回归筛选出的影响因素结合随机搜索获得的最优参数组合优化XGBoost比其他模型预测精度更高,平均相对误差为1.53%,均方根误差为0.1403 m^(3)/min,希尔不等系数为0.0132,研究结果可为现场瓦斯管理提供参考依据。In order to prevent the gas disasters,and solve the problem of inaccuracy in the process of predicting underground gas emission amount due to many influencing factors,a model based on the Lasso regression and limit gradient boosting(XGBoost)optimized with random search was proposed to predict the gas emission amount.Taking the historical data of gas emission amount from the fully-mechanized mining face of a coal mine in Shenyang as an example,the factors affecting the gas emission amount were comprehensively considered.The Lasso regression was used to extract the feature data with important influence on the gas emission amount as the prediction input.The random search algorithm was used to optimize the four key parameters of the XGBoost model,and the optimal parameters were selected to establish a prediction model,so as to obtain the prediction indexes and analyze and compare the other models.The results showed that the influencing factors selected by Lasso regression combined with the optimal parameters combination optimization XGBoost obtained by random search had higher prediction accuracy than the other models.The average relative error was 1.53%,the root mean square error was 0.1403 m^(3)/min,and the Theil inequality coefficient was 0.0132,which provides a reference for the on-site gas management.

关 键 词：随机搜索算法 XGBoost模型 瓦斯涌出量 Lasso回归 

分 类 号：X936[环境科学与工程—安全科学] TD712[矿业工程—矿井通风与安全]