Research Of Coal Mine Gas Gush Predictor Based On Svm

Coal mine gas is one of the greast threats which have influence on the safty of coal mine work. Under a certain concentration of gas and conditions, the gas in coal mine may explode or extrude and such big accidents will happen. In order to prevent these bad accidents happening, not only the gas surveillance should be taken but also the real-time prediction of gas gush should be put into practice. The aim of this paper is trying to research a kind of coal mine gas gush predictor on the platform of embedded computer system, which can predict the relative value and trend of gas gush in one work lane under coal mine. The predictor adopt support vector machine (SVM) as prediction method, compared with traditional prediction methods of coal mine gas gush such as mine statistical method and plot headstream method etc, and other methods based on fractaltheory or neural network have been proposed recently, the SVM prediction method is a kind of limited samples machine learning algorithm and based on statistical learning theory and structure risk minimization. It overcomes the problems of over study, high dimension and local minimum, also has better generalization. Furthermore, it not only be the same with such nor-linear dynamics process like coal mine gas gash but also ensures that the boundary values are global optimization.The first work of this paper is to analyze the various factors that affect gas gush combine with coal mine geognosy and background of coal mine practical work. Then review the basic theory of support vector machine and emphasize the algorithm of nor-linear regression SVM, and come to the practical problem of gas gush prediction the gas gush predict method based on nor-linear regression SVM algorithm is proposed. Then study the design of predictor, the design and exploration of predictor are based on embedded computer system, because of its many advantages such as high compute, low power, small size and high efficiency. The basic functions of predictor are signals collection, SVM model training, and the prediction of samples.The key algorithm of predictor is nor-linear regression SVM and the experiment platform is embedded system platform. It can set several types of SVM model and kernel on predictor, so that we could observe the speed and precision of prediction under different kinds of SVM or kernel.