基于BP神经网络和随机森林的煤中CO_(2)饱和吸附量预测研究  

Prediction of CO_(2)Saturation Adsorption in Coal Based on BP Neural Network and Random Forest

作　　者：侯渊 高飞[2] HOU Yuan;GAO Fei(Shanxi Jinshen Energy Co.,Ltd.,Xinzhou 036500,China;College of Safety Science and Engineering,Liaoning Technical University,Huludao 125100,China)

机构地区：[1]山西省晋神能源有限公司,山西忻州036500 [2]辽宁工程技术大学安全科学与工程学院,辽宁葫芦岛125100

出　　处：《山西煤炭》2023年第4期94-102,共9页Shanxi Coal

摘　　要：为了定量预测采空区不同理化条件下煤对CO_(2)的吸附封存能力,以常温常压下3种煤在不同矿物质含量、含水率和CO_(2)体积分数下的饱和吸附量数据为基础,分别采用机器学习方法中的误差反向传播(BP)神经网络和随机森林算法对实验数据进行训练和预测,对两种模型的预测性能进行了分析和比较,并通过随机森林算法计算了各影响因素的重要性权重。结果表明:BP神经网络和随机森林模型均可以很好地预测煤对CO_(2)的吸附能力,其中BP神经网络模型测试集的平均绝对误差(MAE)和均方根误差(RMSE)分别为0.0225和0.0718,随机森林模型测试集的MAE和RMSE分别为0.0320和0.0905,均小于0.1;BP神经网络模型预测结果中,相对误差小于5%的数据为100%,而随机森林模型预测结果中,相对误差小于5%的数据为82%,且BP神经网络预测模型的平均绝对误差、均方误差和均方根误差均小于随机森林模型,证明BP神经网络在CO_(2)饱和吸附量的预测应用上准确性更高,预测效果更好。CO_(2)体积分数是所有影响吸附的因素中影响程度最大的,约为30%;其次是比表面积,约为26%;CO_(2)体积分数、比表面积和含水率3个因素的重要性权重之和可达75%,远高于其他3个因素;孔体积与平均孔径的影响程度较低,而矿物质含量对吸附量的影响不大。研究结果可为采空区封存CO_(2)技术的应用提供理论支撑,对温室气体的减排具有重要意义。To quantitatively predict the CO_(2)adsorption and storage capacity of coal under different physiochemical conditions in the goaf,based on the saturated adsorption capacity data of three kinds of coal under different mineral contents,water contents,and CO_(2)concentration at ambient temperature and pressure,the error back propagation(BP)neural network and random forest algorithm in machine learning method were used to train and predict the experimental data respectively.The prediction performance of the two models was analyzed and compared,and the importance weight of each influencing factor was calculated by the random forest algorithm.The results show that both the BP neural network and the random forest model can predict the adsorption capacity of coal to CO_(2).The mean absolute error(MAE)and root mean square error(RMSE)of the BP neural network model test set are 0.0225 and 0.0718,respectively.The MAE and RMSE of the random forest model test set are 0.0320 and 0.0905,respectively,both less than 0.1.In the prediction results of the BP neural network model,the data with a relative error of less than 5%are 100%,while in the prediction results of the random forest model,the data with a relative error of less than 5%are 82%.The MAE,mean square error,and RMSE of the BP neural network prediction model are less than those of the random forest model,which proves that the BP neural network has higher accuracy and better prediction effect in the prediction of CO_(2)saturation adsorption.CO_(2)concentration is the most influential factor among all the factors affecting adsorption,about 30%,followed by the specific surface area,about 26%.The sum of the importance weights of CO_(2)concentration,specific surface area,and water content could reach 75%,which is much higher than the other three factors.The influence of pore volume and average pore size is low,while the mineral content has little effect on the adsorption capacity.The research results can provide theoretical support for the application of CO_(2)storage technolog

关 键 词：BP神经网络 随机森林 CO_(2)饱和吸附量 

分 类 号：TD82[矿业工程—煤矿开采]