基于随机森林回归模型的煤胶质层指数预测  

Prediction of coal colloid layer index based on random forest regression model

作　　者：王利民 朱立江 刘金鸽 WANG Limin;ZHU Lijiang;LIU Jinge(National Energy Group Coal Coking Co.,Ltd.,Wuhai,Inner Mongolia 016000,China;Huatai Jiguang Optoelectronic Technology Co.,Ltd.,Shanghai 200091,China)

机构地区：[1]国家能源集团煤焦化有限责任公司,内蒙古乌海016000 [2]华太极光光电技术有限公司,上海200091

出　　处：《中国科技论文》2025年第3期267-276,共10页China Sciencepaper

摘　　要：为了提升煤质检测效率,构建了机器学习回归模型,旨在提升煤质分析的效率和精确度。通过分析99种不同煤样的关键工业参数(包括水分、灰分、挥发分、硫分、粘结指数和胶质层指数),该模型构建了一个全面的数据集。首先,运用皮尔逊相关性分析来识别与胶质层指数密切相关的特征,并采用递归特征消除(recursive feature elimination,RFE)与交叉验证结合的方法进一步精炼特征选择。通过构建3种机器学习模型——随机森林回归模型、支持向量机回归模型和决策树回归模型完成煤胶质层指数预测。通过网格搜索和交叉验证技术对模型的超参数进行优化,以确保模型达到最佳性能。模型的最终评估结果表明,随机森林回归模型预测精度非常高,决定系数(R²)达到了0.969,均方根误差(root mean squared error,RMSE)为0.171,平均绝对误差(mean absolute error,MAE)为0.108。In order to improve the efficiency of coal quality detection,a machine learning regression model was developed to enhance the efficiency and accuracy of coal quality analysis.A comprehensive dataset was con⁃structed by analyzing key industrial parameters from 99 different coal samples,including moisture,ash content,volatile matter,sulfur content,bonding index,and Gieseler fluidity index.First,Pearson correlation analysis was used to identify features closely related to the Gieseler fluidity index,and the method combined recursive fea⁃ture elimination(RFE)with cross-validation was employed for further refinement of feature selection.Three machine learning models,i.e.,random forest regression model,support vector machine regression model,and decision tree regression model,were constructed to predict the Gieseler fluidity index of coal.The model’s hyperparameters were optimized through grid search and cross-validation techniques to ensure optimal perfor⁃mance.The final evaluation results of the model indicates that the random forest regression model reaches a high prediction accuracy with the coefficient of determination(R²),the root mean squared error(RMSE)and the mean absolute error(MAE)of 0.969,0.171 and 0.108,respectively.

关 键 词：胶质层指数 机器学习 随机森林 回归模型 交叉验证 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]