基于梯度提升回归树算法的煤炭发热量计算  

Calculation of Coal Calorific Value Based on Gradient Boosting Regression Tree Algorithm

作　　者：万国祥[1] WAN Guoxiang(Heidaigou Open-pit Coal Mine of Shenhua Junggar Energy Co.,Ltd.,Ordos,Inner Mongolia 017000)

机构地区：[1]神华准格尔能源有限责任公司黑岱沟露天煤矿,内蒙古鄂尔多斯017000

出　　处：《能源科技》2024年第3期85-89,共5页Energy Science and Technology

摘　　要：煤炭发热量是衡量煤质的关键指标,反映了煤炭充分燃烧时释放的能量。煤炭发热量可通过实验测定和计算途径获取,其中实验方法虽精确却复杂昂贵耗时。在实际应用中,通过多元线性回归估算得出发热量数据,但是这种方法计算的结果准确率较低。鉴于此,提出了一种基于梯度提升回归树(GBRT)的煤炭发热量计算方法,该方法是一种机器学习回归分析方法,能够有效克服多元线性回归模型在处理非线性数据时的局限性。在国际公认的COALQUAL煤质数据库上对提出的模型进行了验证和对比,结果显示:GBRT模型的预测误差(MAE、MSE、RMSE)均小于多元线性回归模型;拟合优度(R2=0.989)大于多元线性回归模型(R2=0.970)。说明GBRT是一种高效、准确的煤炭发热量预测模型,对于煤质评价具有一定的实际意义。The calorific value of coal is a key index to measure coal quality,which reflects the energy released when coal is fully burned.The calorific value of coal can be obtained by experimental measurement and calculation,in which the experimental method is accurate but complex,expensive and time-consuming.In practical application,the calorific value data is estimated by multiple linear regression,but the accuracy of this method is low.In view of this,a calculation method of coal calorific value based on gradient boosting regression tree(GBRT)is proposed,which is a machine learning regression analysis method and can effectively overcome the limitations of multiple linear regression models in dealing with nonlinear data.The proposed model is verified and compared on the internationally recognized COALQUAL coal quality database.The results show that the prediction errors(MAE,MSE,RMSE)of GBRT model are less than those of multiple linear regression model,and the fitting goodness(R2=0.989)is greater than that of multiple linear regression model(R2=0.970),demonstrating that GBRT is an efficient and accurate prediction model of coal calorific value and has certain practical significance for coal quality evaluation.

关 键 词：煤炭发热量 梯度提升回归树 回归分析 预测 

分 类 号：TQ533.4[化学工程—煤化学工程]