Research And Application Of Gas Emission Prediction Based On LS-SVM

Gas emission prediction has always been a significant subject in coal mine gas study field. It can provide academic basis for formulating production targets, preventing and dealing with gas accidents. The accuracy of prediction results will directly influence mine's technological and economic targets. Aiming at solving machine learning of limited data samples, the dissertation introduces the least square support vector machine into gas emission prediction based on the analysis of disadvantages of traditional gas emission prediction methods. The study shows that LS-SVM theory gets a higher accuracy than traditional methods.The dissertation selects impact factors of extracting seam, neighboring seam and goaf in mining face according to different-source prediction theory. Extracting seam influence factors are extracting seam gas content, seam thickness, seam dip angle, burial depth, working face length, drawing speed, mining height and recovery ratio. Neighboring seam influence factors are neighboring seam gas content, neighboring seam thickness, interlamination distance, interlamination lithological characters, seam dip angle, drawing speed, mining height, working face length and roof management style. Gob gas emission impact factors are extracting seam gas content, extracting seam thickness, recovery ratio and mining intensity.A gas emission prediction model is built based on the least square support vector machine with Gaussian kernel function whose parameters are optimized by proposed self-tuning mesh searching method. In order to examine the new prediction model's performance, the dissertation selects two other prediction methods, different-source prediction method and backpropagation neural network to make a case study and compares prediction results among them. The results show that LS-SVM prediction model's mean errors are 2.12%,7.7% and 3.38% respectively, while BP neural network prediction model's mean errors are 5.6519%, 8.4744% and 8.6009% respectively, and extracting seam gas emission prediction's mean error based on different-source is 10.4012%. The conclusion that the prediction precision is obviously higher than that of BP neural network and different-source prediction method proves feasibility and superiority to apply LS-SVM theory for gas emission prediction. At last, a software for gas emission prediction based on LS-SVM is realized with MATLAB GUIDE. The software has a friendly interface and puts a positive effect on the wide application of LS-SVM theory.