煤自燃灾害是影响煤矿安全生产的重要因素之一。本文针对传统的煤自燃危险程度预测方法在指标选取、权重赋值方面的特点和存在的问题，考虑煤自燃过程中温度和气体的非线性关系，根据机器学习相关理论构建了基于GA-MK-SVM的煤自燃危险程度预测方法。

通过分析柴家沟矿煤自然发火实验过程中煤温和气体浓度的变化规律发现，柴家沟矿煤样实验最短自然发火期近似为31天。基于煤自燃特性参数分析，确定柴家沟矿煤样临界温度为70℃，干裂温度为100℃。根据升温过程中各气体的变化情况，将CO、CO₂、CH₄、C₂H₆、C₂H₄、CO/CO₂比值、C₂H₄/C₂H₆比值、C₂H₆/CH₄比值、O₂/（CO+CO₂）比值作为煤自燃危险程度的表征指标。基于主成分分析法（PCA）建立了以各表征指标为自变量、以煤温为因变量的煤自燃危险程度预测指标体系。

根据支持向量机（SVM）相关理论，构建了单核函数SVM模型（Poly-SVM和RBF-SVM）和基于多项式核函数和径向基核函数线性组合的混合核函数SVM模型（MK-SVM）。并以柴家沟矿煤自然发火实验为基础，对比各模型的预测效果发现，在未进行参数优化的情况下，MK-SVM模型的预测效果略优于单核函数SVM模型，但预测精度有待进一步提升。

利用粒子群优化算法（PSO）、网格搜索算法（GS）、遗传算法（GA）分别对单核函数SVM模型和MK-SVM模型进行超参数寻优。经参数优化后，各模型的预测精度均有明显提升。其中，GA-MK-SVM模型的预测效果最佳，对应的预测误差温度小于1℃。通过与分类回归树（CART）、随机森林（RF）、多元线性回归（MLR）的横向对比发现，GA-MK-SVM模型的预测精度仍最高，因此确定以GA-MK-SVM模型为煤自燃危险程度预测最优模型。最后，将基于GA-MK-SVM的煤自燃危险程度预测方法应用于大佛寺矿煤自然发火实验数据和采空区现场观测数据，对应的预测误差温度均小于1℃，验证了预测方法具有良好的应用性和普适性。

Coal spontaneous combustion disaster is an important factor affecting coal mine safety. Based on the characteristics and existing problems of index selection and weight assignment of traditional coal spontaneous combustion risk prediction methods, this paper considers the non-linear relationship between temperature and gas in the process of coal spontaneous combustion. The GA-MK-SVM prediction method of the risk of coal spontaneous combustion is constructed according to the related theory of machine learning.

According to the law of temperature and gas concentration in the coal spontaneous combustion experiment, it is found that the shortest spontaneous ignition period of the coal sample in Chaijiagou Mine is approximately 31 days. Based on the analysis of coal spontaneous combustion characteristics, the critical temperature of the Chaijiagou coal sample is 70℃, and the dry cracking temperature is 100℃. According to the changes of each gas during the heating process, CO, CO₂, CH₄, C₂H₆, C₂H₄, CO/CO₂ ratio, C₂H₄/C₂H₆ ratio, C₂H₆/CH₄ ratio, O₂/(CO+CO₂) ratio are regarded as the risk of coal spontaneous combustion characterization index. Based on the Principal Component Analysis (PCA), an index system for predicting the risk of coal spontaneous combustion is established, which takes each indicator as an independent variable and coal temperature as a dependent variable.

According to the related theories of Support Vector Machine (SVM), single-kernel SVM models (Poly-SVM and RBF-SVM) and mixed-kernel SVM model (MK-SVM) which based on the linear combination of polynomial kernel function and radial basis kernel function are constructed. By comparing the effects of the model in predicting the risk of spontaneous combustion of coal in Chaijiagou Mine, it can be concluded that without parameter optimization, the prediction accuracy of the MK-SVM model is better than that of the single-kernel SVM model. However, the prediction accuracy yet to be further improved.

The Particle Swarm Optimization Algorithm (PSO), the Grid Search Algorithm (GS), and the Genetic Algorithm (GA) are respectively used to optimize the hyperparameters of the single-kernel SVM models and the MK-SVM model. The results show that after parameter optimization, the accuracy of each prediction model has been considerably improved. Among them, the GA-MK-SVM model has the highest prediction accuracy, and the corresponding prediction error temperature is less than 1°C. Through horizontal comparison with Classification and Regression Tree (CART), Random Forest (RF), and Multiple Linear Regression (MLR), it is found that the prediction accuracy of GA-MK-SVM model is still highest. Therefore, the GA-MK-SVM model is determined as the optimal model for predicting the risk of coal spontaneous combustion. Finally, applied the coal spontaneous combustion risk prediction method based on the GA-MK-SVM model to the experimental data and field observation data of Dafosi Mine. The results show that the corresponding prediction error temperatures are all less than 1℃, which verifies that the prediction method has good applicability and universality.

[1] 李辉, 徐美宵, 张泉. 改革开放40年中国能源政策回顾:从结构到逻辑[J]. 中国人口·资源与环境, 2019, 29(10): 167-176.

[2] 中华人民共和国自然资源部. 中国矿产资源报告2020[M]. 北京: 地质出版社, 2020: 1-10.

[3] 王国法. 基于新一代信息技术的现代能源治理体系建设研究[J]. 煤炭经济研究, 2020, 40(11): 4-9.

[4] 谢和平, 王金华, 鞠杨, 等. 煤炭革命的战略与方向[M]. 北京: 科学出版社, 2018: 1-31.

[5] 谢和平, 吴立新, 郑德志. 2025年中国能源消费及煤炭需求预测[J]. 煤炭学报, 2019, 44(07): 1949-1960.

[6] 王德明. 煤矿热动力灾害及特性[J]. 煤炭学报, 2018, 43(01): 137-142.

[7] Onifade M, Genc B. A review of research on spontaneous combustion of coal[J]. International Journal of Mining Science and Technology, 2020, 30(03): 303-311.

[8] Beamish BB, Theiler J. Coal spontaneous combustion: Examples of the self-heating incubation process[J]. International Journal of Coal Geology, 2019, 215: 103297.

[9] Zubíček V, Adamus A. Susceptibility of coal to spontaneous combustion verified by modified adiabatic method under conditions of Ostrava–Karvina Coalfield, Czech Republic[J]. Fuel Processing Technology, 2013, 113(113): 63-66.

[10] 王凯. 陕北侏罗纪煤氧化自燃特性实验研究[D]. 西安科技大学, 2013.

[11] 陆新晓, 赵鸿儒, 朱红青, 等. 氧化煤复燃过程自燃倾向性特征规律[J]. 煤炭学报, 2018, 43(10): 2809-2816.

[12] 王凯, 和运中, 尚博. 预氧化对煤复燃极限参数影响的实验研究[J]. 煤矿安全, 2020, 51(07): 31-35.

[13] Wang K, Liu XR, Deng J, et al. Effects of pre-oxidation temperature on coal secondary spontaneous combustion[J]. Journal of Thermal Analysis and Calorimetry, 2019, 138(2): 1363-1370.

[14] 邓军, 何勇军, 翟小伟, 等.基于控制论的八宝煤矿瓦斯爆炸事故分析[J]. 煤矿安全, 2015, 46(11): 234-237.

[15] 郭军, 蔡国斌, 金彦, 等. 煤自燃火灾防治技术研究进展及趋势[J]. 煤矿安全, 2020, 51(11): 180-184.

[16] 邓军, 白祖锦, 肖旸. 离子液体抑制煤自燃的研究进展[J]. 煤矿安全, 2017, 48(10): 39-42+46.

[17] 翟小伟, 成倬, 徐启飞, 等. 粒径对煤低温氧化阶段表观活化能影响试验研究[J]. 煤炭工程, 2020, 52(10): 143-148.

[18] 翟小伟. 煤氧化过程CO产生机理及安全指标研究[D]. 西安科技大学, 2012.

[19] 王福生, 王建涛, 顾亮, 等. 煤自燃预测预报多参数指标体系研究[J]. 中国安全生产科学技术, 2018, 14(06): 45-51.

[20] 邓军, 李贝, 王凯, 等. 我国煤火灾害防治技术研究现状及展望[J]. 煤炭科学技 术, 2016, 44(10): 1-7+101.

[21] 蓝航, 陈东科, 毛德兵. 我国煤矿深部开采现状及灾害防治分析[J]. 煤炭科学技术, 2016, 44(01): 39-46.

[22] 昝军才, 魏成才, 蒋可娟, 等. 基于BP神经网络的煤自燃温度预测研究[J]. 煤炭工程, 2019, 51(10): 113-117.

[23] 邓军, 徐精彩, 张迎弟, 等. 煤最短自然发火期实验及数值分析[J]. 煤炭学报, 1999(03): 52-56.

[24] 文虎. 煤自燃过程的实验及数值模拟研究[D]. 西安科技大学, 2003.

[25] 余明高, 黄之聪, 岳超平. 煤最短自然发火期解算数学模型[J]. 煤炭学报, 2001(05): 516-519.

[26] 余明高, 王清安, 范维澄, 等. 煤层自然发火期预测的研究[J]. 中国矿业大学学报, 2001(04): 64-67.

[27] 邓军, 张燕妮, 徐通模, 等. 煤自然发火期预测模型研究[J]. 煤炭学报, 2004(05): 568-571.

[28] Yang YL, Li ZH, Hou SR, et al. The shortest period of coal spontaneous combustion on the basis of oxidative heat release intensity[J]. International Journal of Mining Science and Technology, 2014, 24(01): 99-103.

[29] 康文杰, 周亮宇, 王德明, 等. 煤最短自然发火期快速预测方法[J]. 工矿自动化, 2019, 45(05): 31-34.

[30] 梁运涛, 宋双林, 罗海珠, 等. 煤自然发火期计算模型及其解析解[J]. 煤炭学报, 2015, 40(09): 2110-2116.

[31] 张辛亥, 席光. 用于煤自然发火期预测的神经网络模型和实验技术[J]. 西安交通大学学报, 2006(09): 1058-1061.

[32] 王华. 煤自然发火期的主成分神经网络预测模型[J]. 计算机工程与应用, 2011, 47(26): 242-245.

[33] 边冰, 刘奇, 何浩. 基于LVQ神经网络的煤自然发火预测系统[J]. 华北理工大学学报(自然科学版), 2017, 39(01): 12-16.

[34] Zhang D, Yang X, Deng J, et al. Research on coal spontaneous combustion period based on pure oxygen adiabatic oxidation experiment[J]. Fuel, 2020, 288: 119651.

[35] 岳宁芳, 金彦, 孙明福, 等. 基于多指标气体的煤自燃进程分级预警研究[J]. 安全与环境学报, 2020, 20(06): 2139-2146.

[36] Guo J, Wen H, Zheng XZ, et al. A method for evaluating the spontaneous combustion of coal by monitoring various gases[J]. Process Safety and Environmental Protection, 2019, 126: 223-231.

[37] Liu HW, Wang F, Ren T, et al. Influence of methane on the prediction index gases of coal spontaneous combustion: A case study in Xishan coalfield, China[J]. Fuel, 2021, 289: 119852.

[38] 翟小伟, 成倬. 柴家沟矿4~(-2)煤层自燃标志气体优选实验研究[J]. 煤矿安全, 2019, 50(11): 18-23.

[39] 武福生. 预测煤自燃的复合气体指标优选实验研究[J]. 工矿自动化, 2018, 44(07): 61-65.

[40] Bi HB, Wang CX, Lin QZ, et al. Combustion behavior, kinetics, gas emission characteristics and artificial neural network modeling of coal gangue and biomass via TG-FTIR[J]. Energy, 2020, 213: 118790.

[41] 王磊, 武术静, 李长青. 基于修正灰色马尔科夫模型的煤自燃预测[J]. 计算机仿真, 2014, 31(11): 416-420.

[42] 张红芬, 陈孝国, 孙显龙, 等. 基于模糊综合评价的煤自燃等级分类研究[J]. 煤炭技术, 2014, 33(12): 209-210.

[43] Wang HY, Fang XY, Li YC, et al. Research and application of the underground fire detection technology based on multi-dimensional data fusion[J]. Tunnelling and Underground Space Technology, 2021, 109:103753.

[44] 刘方园, 王水花, 张煜东. 支持向量机模型与应用综述[J]. 计算机系统应用, 2018, 27(04): 1-9.

[45] Jiang F, Lu Y, Chen Y, et al. Image recognition of four rice leaf diseases based on deep learning and support vector machine[J]. Computers and Electronics in Agriculture, 2020, 179(02): 105824.

[46] 张松兰. 支持向量机的算法及应用综述[J]. 江苏理工学院学报, 2016, 22(02): 14-17+21.

[47] Song XX, Jian L, Song YJ. A chunk updating LS-SVMs based on block Gaussian elimination method[J]. Applied Soft Computing, 2016, 51: 96-104.

[48] 蒋卓芸. 经验模态分解在高光谱遥感数据处理中的应用[D]. 成都理工大学, 2017.

[49] 耿蒲龙, 宋建成, 赵钰, 等.基于SVM增量学习算法的煤矿高压断路器故障模式识别方法[J]. 煤炭学报, 2017, 42(08): 2198-2204.

[50] 张春艳, 倪世宏, 查翔. 一种基于近邻边界的粒度支持向量机学习策略[J]. 计算机科学, 2016, 43(03): 271-274.

[51] 顾吉峰, 王蓓. 基于改进粒子群算法的孪生支持向量机[J]. 计算机工程与设计, 2020, 41(11): 3078-3082.

[52] 梁礼明, 钟震, 陈召阳. 支持向量机核函数选择研究与仿真[J]. 计算机工程与科学, 2015, 37(06): 1135-1141.

[53] 奉国和. SVM分类核函数及参数选择比较[J]. 计算机工程与应用, 2011, 47(03): 123-124.

[54] 施皓晨, 肖海鹏, 周建江. 一种双线性分段二分网格搜索SVM最优参数方法[J]. 计算机与数字工程, 2020, 48(09): 2179-2184.

[55] 黄俊, 刘小生. 基于GS-PSO-SVM模型的边坡稳定性预测模型[J]. 中国矿业, 2020, 29(06): 87-91.

[56] Zhou T, Lu HL, Wang WW, et al. GA-SVM based feature selection and parameter optimization in hospitalization expense modeling[J]. Applied Soft Computing, 2019, 75: 323-332.

[57] Jiang W, Xie YJ, Li WX, et al. Prediction of the splitting tensile strength of the bonding interface by combining the support vector machine with the particle swarm optimization algorithm[J]. Engineering Structures, 2021, 230(07): 111696.

[58] 丁满, 张寿宇, 黄晓光, 等. 基于支持向量机回归与模拟退火算法的产品外观意象设计[J]. 机械设计, 2020, 37(03): 135-140.

[59] 王杨, 刘蒙, 闫伟光. 蚁群优化算法优化支持向量机的视频分类[J]. 现代电子技术, 2020, 43(01): 56-58+62.

[60] 孟倩, 王洪权, 王永胜, 等. 煤自燃极限参数的支持向量机预测模型[J]. 煤炭学报, 2009, 34(11): 1489-1493.

[61] 高原, 覃木广, 李明建. 基于支持向量机的采空区遗煤自燃预测分析[J]. 煤炭科学技术, 2010, 38(02): 50-54.

[62] 方刚, 郭佐宁, 迪明, 等. 基于粗糙集和支持向量机的采空区煤自燃火灾预报[J]. 西安科技大学学报, 2012, 32(06): 712-717.

[63] 邓军, 周少柳, 马砺, 等. 基于PCA-PSOSVM的煤自燃程度预测研究[J]. 矿业安全与环保, 2016, 43(05): 27-31.

[64] Lei CK, Deng J, Cao K, et al. A random forest approach for predicting coal spontaneous combustion [J]. Fuel, 2018, 223: 63-73.

[65] Lei CK, Deng J, Cao K, et al. A comparison of random forest and support vector machine approaches to predict coal spontaneous combustion in gob [J]. Fuel, 2019, 239: 293-311.

[66] 张天宇, 鲁义, 施式亮, 等. 基于支持向量机分类算法的多煤种煤自燃危险性预测[J]. 湖南科技大学学报(自然科学版), 2019, 34(02): 11-17.

[67] 肖旸, 王振平, 马砺, 等. 煤自燃指标气体与特征温度的对应关系研究[J]. 煤炭科学技术, 2008(06): 47-51.

[68] 徐精彩, 薛韩玲, 文虎, 等. 煤氧复合热效应的影响因素分析[J]. 中国安全科学学报, 2001(02): 34-39.

[69] 徐精彩. 自燃危险区域判定理论[M]. 北京: 煤炭工业出版社, 2001.

[70] 陆伟, 胡千庭, 仲晓星, 等.煤自燃逐步自活化反应理论[J]. 中国矿业大学学报, 2007(01): 111-115.

[71] 马砺, 雷昌奎, 王凯, 等. 高地温环境对煤自燃危险性影响试验研究[J]. 煤炭科学技术, 2016, 44(01): 144-148+156.

[72] 翟小伟, 蒋上荣, 王博. 水分对煤孔隙结构及自燃特性的影响研究现状[J]. 煤矿安全, 2020, 51(02): 38-42.

[73] 余明高, 袁壮, 褚廷湘, 等. 不同自燃性煤氧化阶段的表征差异[J]. 重庆大学学报, 2017, 40(02): 37-44.

[74] 谭波, 邵壮壮, 郭岩, 等. 基于指标气体关联分析的煤自燃分级预警研究[J]. 中国安全科学学报, 2021, 31(02): 33-39.

[75] 韩亮. 煤自燃指标气体生成特性及内在规律性研究[D]. 太原理工大学, 2017.

[76] 王汉元, 贾宝山, 李守国, 等. 基于PCA的煤炭自然发火预测预报研究[J]. 洁净煤技术, 2015, 21(06): 119-122.

[77] Deng JC, Ge SK, Qi HN, et al. Underground coal fire emission of spontaneous combustion, Sandaoba coalfield in Xinjiang, China: Investigation and analysis [J]. Science of The Total Environment, 2021, 777: 146080.

[78] 徐雅静, 汪远征. 主成分分析应用方法的改进[J]. 数学的实践与认识, 2006(06): 68-75.

[79] Vapnik VN. An Overview of Statistical Learning Theory [J]. IEEE Transactions on Neural Networks. 1999, 10: 988-999.

[80] Gaspar P, Carbonell J, Oliveira, JL. On the parameter optimization of Support Vector Machines for binary classification[J]. Journal of integrative bioinformatics, 2012, 9(03): 33-43.

[81] Taghavifar H, Mardani A. A comparative trend in forecasting ability of artificial neural networks and regressive support vector machine methodologies for energy dissipation modeling of off-road vehicles[J]. Energy, 2014, 66: 569-576.

[82] 张倩, 杨耀权. 基于支持向量机核函数的研究[J]. 电力科学与工程, 2012, 28(05): 42-45.

[83] 杨维, 李歧强. 粒子群优化算法综述[J]. 中国工程科学, 2004(05): 87-94

[84] 马永杰, 云文霞. 遗传算法研究进展[J]. 计算机应用研究, 2012, 29(04): 1201-1206+1210.

[85] Zhong KQ, Xiao Y, Zhao X, et al. Predictive ability of four statistical models for determining the influence of coal thermophysical properties during the initial phase of coal spontaneous combustion[J]. Fuel, 2021, 292(01): 120348.

[86] 刘春卫, 罗健旭. 基于混合核函数的PSO-SVM分类算法[J]. 华东理工大学学报(自然科学版), 2014, 40(01): 96-101.

[87] 李方伟, 罗嘉, 朱江, 等. 一种基于混合核函数PSO_SVR的网络安全态势预测方法[J]. 微电子学与计算机, 2015, 32(12): 110-115.

[88] 薛红军, 陈广交, 李鑫民, 等. 基于决策树理论的交通流参数短时预测[J]. 交通信息与安全, 2016, 34(03): 64-71.

[89] Breiman LI, Friedman JH, Olshen RA, et al. Classification and regression trees[J]. Encyclopedia of Ecology, 2015, 57(3): 582-588.

[90] 王宏, 张强, 王颖, 等. 基于ELM的改进CART决策树回归算法[J]. 计算机系统应用, 2021, 30(2): 201-206.

[91] Breiman L. Random forests[J]. Machine Learning, 2001,45(01): 5-32.

[92] 王福生, 张志明, 武建国, 等. 煤体结构对自燃倾向性影响研究[J]. 煤炭科学技术, 2020, 48(05): 83-88.

[93] 曹凯. 综放采空区遗煤自然发火规律及高效防治技术[D]. 中国矿业大学, 2013.