Prediction method of coal spontaneous combustion based on relevance vector machine

Abstract

In terms of coal spontaneous combustion degree prediction, the radial basis function (RBF) neural network method is complex in structure and easy to fall into local optimum, the kernel function based on support vector machine (SVM) is sensitive to parameters due to Mercer condition, the traditional machine learning method has a large error. For the above problems, a coal spontaneous combustion prediction method based on relevance vector machine (RVM) is proposed. Taking Tingnan Coal Mine which is prone to spontaneous combustion as an example, the temperature rising process of coal sample spontaneous combustion is simulated, and the data of gas concentration and coal spontaneous combustion temperature are collected to establish training samples and test samples. The RVM model is constructed from the training samples, and the optimal parameters of the model are obtained. The test samples are substituted into the trained RVM model to predict coal spontaneous combustion temperature. Compared with coal spontaneous combustion prediction methods based on RBF neural network and SVM, the results show that the coal spontaneous combustion prediction methods based on RBF neural network and SVM have small training error but large test error, which indicates that the two methods have over fitting phenomenon and poor generalization ability. The training error and test error of the coal spontaneous combustion prediction method based on RVM are close and prediction accuracy is the highest.

Keywords

• coal spontaneous combustion
• gas concentration
• coal spontaneous combustion temperature
• relevance vector machine
• rvm