富油煤可通过中低温热解得到半焦、焦油和煤气，是集煤、油、气属性于一体的特殊煤炭资源。焦油产率是富油煤低温干馏利用最重要的煤质参数，决定着不同级别煤炭资源的清洁利用的方向。然而由于以往对富油煤的认识不足，在煤炭地质勘查阶段对煤的焦油产率的测试数据十分有限且难以补测，严重制约着富油煤资源的精细评价和高效利用。为了弥补煤焦油产率测试数据的不足，本研究以陕北侏罗纪煤田神府矿区煤质测试数据为支撑，通过各煤质数据与焦油产率之间的相关关系的挖掘，利用BP神经网络和随机森林两种机器学习算法建立了多煤质指标下的焦油产率预测模型，并对预测模型的准确性和泛化性能进行了分析和评价。研究取得的成果如下：

（1）通过对煤的工业分析、元素分析和灰成分分析测试数据与煤焦油产率相关关系分析，结合地质分析和数学关系分析确定了M_ad、A_d、V_daf、C_daf、H_daf、N_daf、O_daf、S_t,d和Al₂O₃与焦油产率相关性较为密切，并选用为焦油产率预测问题的特征参量。

（2）将收集的上述煤质指标齐全的163组数据按照9:1的比例分为训练集数据和测试集数据，利用训练集分别训练建立了BP神经网络焦油产率预测模型和随机森林焦油产率预测模型。在训练集数据上，BP神经网络模型预测值与实际值之间的均方误差为0.30，随机森林模型预测值与实际值之间的均方误差为0.44。利用测试集数据测试了两模型对焦油产率的预测精度，结果显示BP神经网络模型和随机森林模型预测焦油产率的平均绝对误差分别为0.59和0.53，可决系数分别为0.76和0.79。在测试集上，随机森林模型的预测效果略优于BP神经网络模型。

（3）采用SHAP算法探讨了两模型中各个煤质指标的重要性以及它们对焦油产率预测结果的影响。两模型都反映出在预测中挥发分、硫分、碳元素是焦油产率的正向影响因素，氧化铝含量、水分产率是焦油产率的负向影响因素；随机森林模型比BP神经网络模型较好的体现了氢元素和氮元素含量的正向作用，是其预测效果略优的原因。

Tar-rich coal can be pyrolyzed at medium and low temperatures to obtain semi-coke, tar and gas. It is a special coal resource that integrates coal, oil and gas properties. The tar yield of coal is the most important coal quality parameter for the low-temperature carbonization of tar-rich coal, which determines the direction of clean utilization of different levels of coal resources. Due to the lack of understanding of tar-rich coal in the past, the test data of coal tar yield in the stage of coal geological exploration is very limited, which seriously restricts the fine evaluation and efficient utilization of tar-rich coal resources. In order to make up for the deficiency of coal tar yield test data, this study is based on the previous coal quality test data of Shenfu mining area in Jurassic coalfield in northern Shaanxi. Through the mining of the correlation between conventional coal quality data and tar yield, two machine learning algorithms of BP neural network and random forest are used to establish the prediction model of tar yield under multiple coal quality indexes, and the accuracy and generalization performance of the prediction model are analyzed and evaluated. The results of this study are as follows:

（1）Through the analysis of the correlation between the test data of proximate analysis, elemental analysis and ash composition analysis of coal and the yield of coal tar, combined with geological experience and mathematical calculation, it is determined that M_ad, A_d, V_daf, C_daf, H_daf, N_daf, O_daf, S_t,d and Al₂O₃ are closely related to the yield of tar, and are selected as the characteristic parameters for the prediction of tar yield.

（2）The collected 163 sets of data with complete coal quality indexes were divided into training set data and test set data according to the ratio of 9:1. The BP neural network tar yield prediction model and the random forest tar yield prediction model were trained and established by using the training set. On the training set data, the mean square error between the predicted value and the actual value of the BP neural network model is 0.30, and the mean square error between the predicted value and the actual value of the random forest model is 0.44. The prediction accuracy of the two models for tar yield was tested by using the test set data. The results showed that the average absolute errors of the BP neural network model and the random forest model for predicting tar yield were 0.59 and 0.53, respectively, and the coefficients of determination were 0.76 and 0.79. On the test set, the prediction effect of the random forest model is slightly better than that of the BP neural network model.

（3）The SHAP algorithm is used to discuss the importance of each coal quality index in the two models and their influence on the prediction results of tar yield: Both models reflect that volatile matter, sulfur and carbon are positive influencing factors of tar yield in the prediction, and alumina content and water yield are negative influencing factors of tar yield. The random forest model reflects the positive effect of hydrogen and nitrogen content higher than the BP neural network model, which may be the reason for its slightly better prediction effect.

[1] 葛世荣, 樊静丽, 刘淑琴, 等. 低碳化现代煤基能源技术体系及开发战略[J]. 煤炭学报, 2024,49(01): 203-223.

[2] 国家统计局. 中国统计年鉴2023[M]. 北京: 中国统计出版社, 2024.

[3] 袁亮. 煤炭工业碳中和发展战略构想[J]. 中国工程科学, 2023: 1-8.

[4] 葛世荣, 王兵, 冯豪豪, 等. 煤基能源动态碳中和模式及其保供降碳效益评估[J]. 中国工程科学, 2023,25(05): 122-135.

[5] 杨宇, 于宏源, 鲁刚, 等. 世界能源百年变局与国家能源安全[J]. 自然资源学报, 2020,35(11): 2803-2820.

[6] 王双明, 王虹, 任世华, 等. 西部地区富油煤开发利用潜力分析和技术体系构想[J]. 中国工程科学, 2022,24(3): 49-57.

[7] 赵文智, 梁坤, 王坤, 等. 油气安全战略与“双碳”战略:关系与路径[J]. 中国科学院院刊, 2023,38(01): 1-10.

[8] 王双明, 刘浪, 赵玉娇, 等. “双碳”目标下赋煤区新能源开发——未来煤矿转型升级新路径[J]. 煤炭科学技术, 2023,51(01): 59-79.

[9] LI G, WANG L, WANG C, et al. Experimental study on coal gasification in a full-scale two-stage entrained-flow gasifier[J]. ENERGIES, 2020,13(18).

[10] 王双明, 申艳军, 宋世杰, 等. “双碳”目标下煤炭能源地位变化与绿色低碳开发[J]. 煤炭学报, 2023,48(07): 2599-2612.

[11] 田瀚, 李宁, 王双明, 等. 富油煤焦油产率测井评价方法研究[J]. 煤田地质与勘探, 2024.

[12] 东振, 张梦媛, 陈艳鹏, 等. 三塘湖-吐哈盆地富油煤赋存特征与资源潜力分析[J]. 煤炭学报, 2023,48(10): 3789-3805.

[13] 矿产资源工业要求手册编委会. 《矿产资源工业要求手册》(2014年修订本)[M]. 北京: 地质出版社, 2014.

[14] 煤炭科学技术研究院有限公司. MT/T 1179-2019 煤中焦油含量分级[S]. 2019.

[15] 王双明, 师庆民, 王生全, 等. 富油煤的油气资源属性与绿色低碳开发[J]. 煤炭学报, 2021,46(5): 1365-1377.

[16] 杨甫, 段中会, 马丽, 等. 陕西省富油煤分布及受控地质因素[J]. 煤炭科学技术, 2023,51(03): 171-181.

[17] 马丽, 王双明, 段中会, 等. 陕西省富油煤资源潜力及开发建议[J]. 煤田地质与勘探, 2022,50(2): 1-8.

[18] 马丽, 段中会, 贺丹, 等. 陕北三叠纪煤田中—特高油产率煤分布规律及成因机理[J]. 中国煤炭地质, 2022,34(01): 32-36.

[19] 谢青, 李宁, 姚征, 等. 黄陵矿区富油煤焦油产率特征及主控地质因素分析[J]. 中国煤炭, 2020,46(11): 83-90.

[20] 许婷, 李宁, 姚征, 等. 陕北榆神矿区富油煤分布规律及形成控制因素[J]. 煤炭科学技术, 2022,50(03): 161-168.

[21] 王锐, 夏玉成, 马丽. 榆神矿区富油煤赋存特征及其沉积环境研究[J]. 煤炭科学技术, 2020,48(12): 192-197.

[22] 李华兵, 姚征, 李宁, 等. 神府矿区5-2煤层富油煤赋存特征及资源潜力评价[J]. 煤田地质与勘探, 2021,49(03): 26-32.

[23] 姚征, 罗乾周, 李宁, 等. 陕北石炭–二叠纪富油煤赋存特征及影响因素[J]. 煤田地质与勘探, 2021,49(03): 50-61.

[24] 周国庆, 姜尧发, 高峰, 等. 陕西省子长矿区晚三叠世5号煤层的煤岩煤质特征[J]. 中国煤炭, 2015,41(01): 30-34.

[25] JIN L, ZHAO H, WANG M, et al. Effect of temperature and simulated coal gas composition on tar production during pyrolysis of a subbituminous coal[J]. FUEL, 2019,241: 1129-1137.

[26] YU W, HAN S, LEI Z, et al. The reaction behavior of volatiles generated from lignite pyrolysis[J]. FUEL, 2019,244: 22-30.

[27] NGUYEN M, BERNDT C, REICHEL D, et al. Pyrolysis behaviour study of a tar- and sulphur-rich brown coal and GC-FID/MS analysis of its tar[J]. JOURNAL OF ANALYTICAL AND APPLIED PYROLYSIS, 2015,115: 194-202.

[28] PAN S, HUI S, LIANG L, et al. Weight Loss and Tar Evolution during Coal Devolatilization at Various Heating Rates[J]. ENERGY & FUELS, 2015,29(2): 520-529.

[29] SHI Q, LI C, WANG S, et al. Effect of the depositional environment on the formation of tar-rich coal : A case study in the northeastern Ordos Basin, China[J]. JOURNAL OF PETROLEUM SCIENCE AND ENGINEERING, 2022,216.

[30] 师庆民, 王双明, 王生全, 等. 神府南部延安组富油煤多源判识规律[J]. 煤炭学报, 2022,47(5): 2057-2066.

[31] 郭晨, 王生全, 师庆民, 等. 神府南部矿区低阶煤化学组成与工艺性质：特征、关系与实践[J]. 煤田地质与勘探, 2021,49(01): 87-99.

[32] 闫和平, 段中会, 王金锋. 黄陵矿区富油煤焦油产率与补偿密度关系模型预测方法研究[J]. 中国煤炭地质, 2022,34(10): 25-30.

[33] QIAO J, WANG C, SU G, et al. Application research on the prediction of tar yield of deep coal seam mining areas based on PSO-BPNN machine learning algorithm[J]. FRONTIERS IN EARTH SCIENCE, 2023,11.

[34] 赵军龙, 闫和平, 王金锋, 等. 基于测井信息的煤焦油产率预测方法研究[J]. 地球物理学进展, 2023,38(04):1702-1712.

[35] 周永章, 王俊, 左仁广, 等. 地质领域机器学习、深度学习及实现语言[J]. 岩石学报, 2018,34(11): 3173-3178.

[36] HINTON G E, OSINDERO S, TEH Y. A fast learning algorithm for deep belief nets[J]. NEURAL COMPUTATION, 2006,18(7): 1527-1554.

[37] 周永章, 左仁广, 刘刚, 等. 数学地球科学跨越发展的十年：大数据、人工智能算法正在改变地质学[J]. 矿物岩石地球化学通报, 2021,40(03): 556-573.

[38] 常力恒, 朱月琴, 张戈一, 等. 面向矿产资源信息的空间关联性分析[J]. 岩石学报, 2018,34(02): 314-318.

[39] 刘心怡, 周永章. 关联规则算法在粤西庞西垌地区元素异常组合研究中的应用[J]. 地学前缘, 2019,26(04): 125-130.

[40] 陈利成, 陈建宏. 基于数据填补-机器学习的煤与瓦斯突出预测效果研究[J]. 中国安全生产科学技术, 2022,18(9): 69-74.

[41] 侯霖莉, 吴松, 易建洲, 等. 基于机器学习的绿泥石微量元素判别矿床类型[J]. 地球科学, 2023: 1-24.[2024-05-26].http://kns.cnki.net/kcms/detail/42.1874.p.20230912.1324.002.html.

[42] 孙洪亮. 煤质检验数据分析及热值预测研究[D]. 吉林: 东北电力大学, 2014.

[43] DINDARLOO S, HOWER J C, BAGHERIEH A, et al. Fundamental evaluation of petrographic effects on coal grindability by seasonal autoregressive integrated moving average (SARIMA)[J]. INTERNATIONAL JOURNAL OF MINERAL PROCESSING, 2016,154: 94-99.

[44] 苗霖田. 榆神府矿区主采煤层赋存规律及煤炭开采对水资源影响分析[D]. 西安: 西安科技大学, 2008.

[45] 康帅直, 穆琪, 赵永华, 等. 黄土高原神府资源开采区生态环境质量时空格局特征[J]. 生态学报, 2023,43(07): 2781-2792.

[46] 陈陇刚, 陈磊, 陈丽荣, 等. 陕北侏罗纪煤田神府矿区含煤地层及煤层的重新厘定划分[J]. 陕西地质, 2005(01): 70-78.

[47] 晋香兰, 张泓. 鄂尔多斯盆地侏罗系成煤系统[J]. 煤炭学报, 2014,39(S1): 191-197.

[48] 高浩. 陕北富油煤热解提油基础特性及煤焦油净化机理研究[D]. 西安: 西安科技大学, 2021.

[49] 张蕾, 韩智坤, 舒浩, 等. 陕北富油煤低温热解提油基础特性[J]. 煤炭工程, 2022,54(09): 124-128.

[50] 王双明, 师庆民, 王生全, 等. 富油煤的油气资源属性与绿色低碳开发[J]. 煤炭学报, 2021,46(05): 1365-1377.

[51] YAN M W. The impact of coal tar yield and properties research[J]. ADVANCED MATERIALS RESEARCH, 2014,3137(912-914).

[52] HU E, ZENG X, MA D, et al. Effect of the moisture content in coal on the pyrolysis behavior in an indirectly heated fixed-bed reactor with internals[J]. ENERGY & FUELS, 2017,31(2).

[53] 刘阳, 刘捷成, 俞海淼, 等. 新型镍基镁渣催化重整松木热解挥发分焦油析出特性研究[J]. 化工学报, 2019,70(08): 2991-2999.

[54] 石晓莉, 陈水渺, 孙宝林, 等. 不同水分褐煤快速热解试验研究[J]. 洁净煤技术, 2018,24(04): 60-64.

[55] GUO H, SHI H, WU Y, et al. Mineral transformation during rapid heating and cooling of Zhundong coal ash[J]. FUEL, 2022,310.

[56] 矫天禹. 矿物质组分对煤流化床热解特性影响的实验研究[D]. 杭州: 浙江大学, 2018.

[57] 李伍, 杨文斌, 战星羽, 等. 煤有机大分子碳结构石墨化机制[J]. 煤炭学报, 2023,48(02): 855-868.

[58] 李祥, 刘翠茹, 李春艳. 煤的元素分析与工业分析关系的逐层分析与模型构建[J]. 中国煤炭, 2017,43(04): 99-104.

[59] RICHARDS A P, JOHNSON C, FLETCHER T H. Correlations of the elemental compositions of primary coal tar and char[J]. ENERGY & FUELS, 2019,33(10): 9520-9537.

[60] 石振, 成帅, 陈兆辉, 等. 神木烟煤热解产物分布及主要元素的迁移规律[J]. 过程工程学报, 2016,16(05): 802-811.

[61] 李梅, 杨俊和, 夏红波, 等. 典型炼焦高硫煤热解过程中硫迁移规律研究[J]. 煤炭转化, 2013,36(04): 41-45.

[62] 张天印, 冯烁, 李鑫, 等. 富油煤分子结构对生烃潜力影响研究[J]. 新疆地质, 2023,41(S1): 61.

[63] GUO H, SHI H, WU Y, et al. Mineral transformation during rapid heating and cooling of Zhundong coal ash[J]. FUEL, 2022,310(PB).

[64] 宋青锋, 肖宪东, 魏晓云, 等. 基于PGNAA技术的煤灰分主要元素的在线检测[J]. 现代矿业, 2019,35(04): 231-232.

[65] 彭志威. 预处理对鄂尔多斯褐煤热解特性的影响[D]. 西安: 西北大学, 2021.

[66] 樊义龙, 王宁波, 徐红东, 等. 低温煤焦油产率和性质影响因素的研究[J]. 洁净煤技术, 2010,16(6): 36-39.

[67] 杨玉坤. 煤灰及其矿物组分对煤热解影响的实验研究[D]. 杭州: 浙江大学, 2017.

[68] 王美君, 杨会民, 何秀风, 等. 铁基矿物质对西部煤热解特性的影响[J]. 中国矿业大学学报, 2010,39(03): 426-430.

[69] 贾永斌, 黄戒介, 程中虎, 等. 煤快速热解过程中氧化钙对焦油裂解的影响[J]. 煤炭转化, 2001,24(2): 53-57.

[70] 贾永斌, 黄戒介, 王洋. 氧化钙在流化床稀相段对焦油裂解的影响[J]. 中国矿业大学学报, 2004,33(5): 552-556.

[71] 李焕同, 朱志蓉, 张战波, 等. 柠条塔煤矿富油煤赋存特征及其成煤环境控制[J]. 煤炭技术, 2022,41(04): 63-66.

[72] 林海飞, 周捷, 金洪伟, 等. 基于特征选择与机器学习的煤与瓦斯突出危险等级协同预测方法[J]. 采矿与安全工程学报, 2023,40(2): 361-370.

[73] 刘高宇. 基于现代神经网络的煤炭质量数据分析与预测[D]. 郑州: 华北水利水电大学, 2020.

[74] KATIPOGLU O M. Predicting hydrological droughts using ERA 5 reanalysis data and wavelet-based soft computing techniques[J]. ENVIRONMENTAL EARTH SCIENCES, 2023,82(24).

[75] NOWAK M, PAWLOWSKA-NOWAK M, KOKOCINSKA M, et al. The evaluation of grey relative incidence[J]. GREY SYSTEMS-THEORY AND APPLICATION, 2023.

[76] 颜丙峰, 王光耀, 朱肖曼, 等. 逐步回归分析中国煤质特性与其液化特性关系[J]. 石油学报(石油加工), 2021,37(02): 362-371.

[77] LI S, CHEN J, LIU C, et al. Mineral prospectivity prediction via convolutional neural networks based on geological big data[J]. JOURNAL OF EARTH SCIENCE, 2021,32(02): 327-347.

[78] HONG S, ZUO R, HUANG X, et al. Distinguishing IOCG and IOA deposits via random forest algorithm based on magnetite composition[J]. JOURNAL OF GEOCHEMICAL EXPLORATION, 2021,230.

[79] RODRIGUEZ-GALIANO V, SANCHEZ-CASTILLO M, CHICA-OLMO M, et al. Machine learning predictive models for mineral prospectivity : An evaluation of neural networks , random forest , regression trees and support vector machines[J]. ORE GEOLOGY REVIEWS, 2015,71: 804-818.

[80] 克里斯·阿尔本. Python机器学习手册[M]. 电子工业出版社, 2019: 362.

[81] 邢晁瑞, 李磊, 王立彬, 等. 华北型煤田底板破坏深度BP神经网络预测模型研究[J]. 矿业科学学报, 2023,8(05): 688-694.

[82] 李星, 高建良, 张学博, 等. BP神经网络预测地表沉降参数的算法与拓扑结构寻优[J]. 中国安全生产科学技术, 2023,19(06): 90-97.

[83] GAO F, WANG P, WANG D, et al. Model for predicting CO2 adsorption in coal left in goaf based on backpropagation neural network[J]. ENERGIES, 2023,16(9).

[84] BAI G, XU T. Coal mine safety evaluation based on machine learning: a bp neural network model[J]. COMPUTATIONAL INTELLIGENCE AND NEUROSCIENCE, 2022,5233845.

[85] LI Q, LI M Q, WANG R. WOA- BP algorithm for crack detection in coalbed methane reservoirs and its application[J]. CHINESE JOURNAL OF GEOPHYSICS-CHINESE EDITION, 2022,65(2): 773-784.

[86] WANG J, HUANG Y. Evaluation of mine water quality based on the PCA-PSO-BP model[J]. JOURNAL OF WATER AND CLIMATE CHANGE, 2023,604

[87] SHAO W, GUAN Q, TAN Z, et al. Application of BP-ANN model in evaluation of soil quality in the arid area, northwest China[J]. SOIL & TILLAGE RESEARCH, 2021,208.

[88] ZHANG J, LI C, ZHANG T. An Assessment of the Mobility of Toxic Elements in Coal Fly Ash Using the Featured BPNN Model[J]. SUSTAINABILITY, 2023,15(23).

[89] MA D, YAO Q, LIU Y, et al. Prediction of volatile composition of coal pyrolysis under various swelling and catalytic conditions based on recurrent neural network[J]. JOURNAL OF ANALYTICAL AND APPLIED PYROLYSIS, 2023,175.

[90] 陈跃, 王丽雅, 李国富, 等. 基于随机森林算法的低煤阶煤层气开发选区预测[J]. 油气藏评价与开发, 2022,12(04): 596-603.

[91] 付君豪, 陈再贺, 付宪, 等. 利用随机森林算法预测束鹿凹陷束21井区馆陶组储层砂地比[J]. 大庆石油地质与开发, 2023,42(06): 34-41.

[92] MA X, CHEN Z, CHEN P, et al. Predicting the utilization factor of blasthole in rock roadways by random forest[J]. UNDERGROUND SPACE, 2023,11: 232-245.

[93] VASIC M V, JANTUNEN H, MIJATOVIC N, et al. Influence of coal ashes on fired clay brick quality: Random forest regression and artificial neural networks modeling[J]. JOURNAL OF CLEANER PRODUCTION, 2023,407.

[94] WANG K, AI Z, ZHAO W, et al. A hybrid model for predicting low oxygen in the return air corner of shallow coal seams using random forests and genetic algorithm[J]. APPLIED SCIENCES-BASEL, 2023,13(4).

[95] 闫长斌, 高子昂, 姚西桐, 等. 考虑不确定性的TBM施工速度加权随机森林预测模型[J]. 岩土工程学报, 2023,45(12): 2575-2583.

[96] ZHOU J, ZHANG Y, LI C, et al. Enhancing the performance of tunnel water inflow prediction using random forest optimized by grey wolf optimizer[J]. EARTH SCIENCE INFORMATICS, 2023,16(3): 2405-2420.

[97] 桂金咏, 李胜军, 高建虎, 等. 基于特征变量扩展的含气饱和度随机森林预测方法[J]. 岩性油气藏, 2024,36(02): 65-75.

[98] 郭骐嘉, 姚宜斌, 周永江. 融合GNSS气象参数的PM2.5随机森林预测模型[J]. 测绘科学, 2021,46(04): 37-42.

[99] 张爱华, 陶明信, 刘朋阳, 等. 煤中氮的赋存状态与含量分布研究进展[J]. 煤田地质与勘探, 2016,44(01): 9-16.

[100] 师庆民, 米奕臣, 王双明, 等. 富油煤热解流体滞留特征及其机制[J]. 煤炭学报, 2022,47(03): 1329-1337.