Research On The Prediction Methods Of Gas Emission Based On Data Mining Techniques

Gas disaster is one of the major disasters of the coal mine, which not only caused thegreat threat to the safety of the staff, but also brought a lot of property damage. It isessential to prevent and control the gas. So, how to predict the gas emission is theimportant part of that and it would be a large impact of production safety. Therefore, it hasgreat significance to select the best method to predict the gas emission according to thedifferent situations in coal mines.It used the theories and methods of data mining in the paper to forecast the gasemission and analysis the reasons. Also, it used Poly Analyst and MATLAB to build andcalculate the model.(1)Correlation analysis and gray relational analysis were used to analyze the gasemission and the impact factors, and it could be known that:①gas emission was into apositive correlation with the amount of gas content of coal seam, depth, seam thickness,mining intensity and the gas content of the adjacent layers; Gas emission was into anegative correlation with advance speed and the out rate of workface.②the absolute valueof the correlation coefficient with the influencing factors and gas emission was over0.71,it was significantly related of highly correlated.③it could be determined that theinfluencing factors were the amount of gas content of coal seam, depth, seam thickness,mining intensity and the gas content of the adjacent layers.(2)Poly Analyst support vector was used to predict the gas emission and twoparameters of kernel function were selected to predict the test data. It could be knowthrough the results:①when the polynomial kernel function parameters and the degree ofthe polynomial was5, the minimum average relative error was0.91%.②when theGaussian kernel function parameters and the standard deviation was2.1, the minimumaverage relative error was8.59%. (3)By predicting the test data which calculated through the two nuclear functions, itcould be known that: the average relative error which predicted by polynomial kernelfunction was3.04%, the average relative error which predicted by Gaussian kernel was5.39%, the former one was better than the latter. It was simple and convenient to predictthe gas emission by Poly Analyst support vector, and the results were good, it could be anew way to predict the gas emission. But the kernel function should be selected moreproperly according to the nature of the object while using the models.(4)The BP neural network was created which meets the requirements of the networkdesign based on the MATLAB. According to the training data and the test data, it could beknown that:①t hough the capability of network mapping wouldbe proved with theincreasing in the hidden layer nodes, the forecast precision would not necessarily improve.②it was accuracy to predict the data; the maximum relative error was8.14%and theaverage relative error was3.68%; all the errors were less than10%.③in order to improvethe network prediction accuracy, the samples data should be more and accurately, but alsothe appropriate convergence error should be determined.