摘要
在对煤自然发火程度预测的方法中,采用径向基(RBF)神经网络预测煤自燃的方法容易发生局部最优、结构冗杂现象,采用支持向量机(SVM)预测会由于Mercer条件的限制导致其核函数对参数反应敏感,常规的机器学习方法存在较大误差,因此引入以相关向量机(RVM)为基础进行预测。结合晋能控股集团四老沟煤矿实际生产情况,对煤自燃升温过程进行数值模拟,并构建样本。以训练样本为基础建立相关向量机(RVM)模型,获得模型最佳参数;在已训练的模型中代入测试样本,对煤自燃温度进行预测。通过对比SVM和RBF预测方法,结果证明:采用SVM和RBF神经网络预测煤自燃的方法虽然训练误差不大,但是测试误差较高,具有“过拟合”问题,泛化能力不高,而采用RVM预测方法,其测试误差和训练误差二者相对接近,而且具有最高的预测精度。
On the degree of coal spontaneous combustion prediction method,the radial basis(RBF)neural network prediction method of coal spontaneous combustion will be prone to local optimum,the structure's mad phenomenon,using support vector machine(SVM)to predict the coal spontaneous combustion method,due to restrictions of the Mercer condition,its kernel function is sensitive to parameters of reaction,conventional methods of machine learning there is a big error.Therefore,the prediction of spontaneous combustion based on correlation vector machine(RVM)is introduced.Combined with the actual production situation of Silaogou Coal Mine of Jineng Holding Group,the numerical simulation of coal spontaneous combustion heating process was carried out,and the data of gas and temperature change generated in the test process were recorded,and samples were constructed.A correlation vector machine(RVM)model was established based on training samples,and the optimal parameters of the model were obtained.By substituting test samples into the trained model,the spontaneous combustion temperature of coal is predicted.Compared with the other two prediction methods(SVM and RBF neural network),the results show that although the training error of SVM and RBF neural network prediction method is not large,but the test error is high,indicating that these two prediction methods have the problem of"over-fitting",and the generalization ability is not high.RVM is used to predict coal spontaneous combustion,and its test error and training error are relatively close,and it has the highest prediction accuracy.
作者
王丹
WANG Dan(Silaogou Coal Mine of Jineng Holding Mining Group, Datong 037000,China)
出处
《煤》
2022年第4期1-5,共5页
Coal
基金
国家自然科学基金青年基金项目“深部开采复杂条件下煤层氧化特性研究”(51304070)。
