Least Squares Support Vector Machine for Gas Concentration Forecasting in Coal Mine

Gas concentration, which is a chaotic time series in essence, is a key factor of the coal mine safety. An accurate forecast of gas concentration is required to guarantee safety and has very highly social and economic benefits. Least squares support vector machine (LS-SVM) has been receiving increasing interest in areas ranging from its original application in pattern recognition to other applications such as regression estimation due to its remarkable generalization performance. In this paper, LS-SVM is a promising method for the forecasting of gas concentration because it uses a risk function consisting of the empirical error and a regularized term which is derived from the structural risk minimization principle. The variability in performance of LSSVM with respect to the free parameters is investigated experimentally. In addition, this study examines the feasibility of applying LS-SVM in gas concentration forecasting by comparing it with the multilayer back-propagation neural network (BPNN) and the regularized radial basis function neural network (RBFNN). The experimental results show that among the three methods, LS-SVM outperforms the BPNN gas concentration forecasting, and there are comparable generalization performance between LS-SVM and RBFNN, but LS-SVM converges faster than the RBFNN. Finally, LS-SVM provides a promising alternative for gas concentration forecasting.