Support Vector Machine And Its Application In Burst Prediction Model

Support Vector Machine (SVM) is used to solve the limited sample learning problems, is not sensitive to the dimensions of the data and variability, and has better classification accuracy and generalization ability. SVM method has been successfully used for isolation of handwriting recognition, text classification, face recognition and so on, and has shown great advantages. SVM is the essence of the problem with two types of classification and regression problem,but many of the issues are multi-classification problems, how to extend SVM to multi-class classification problem has been a hot research.Burst is a complex nonlinear dynamic phenomena, is a kind of induced geological disasters by human activities such as coal mining. Now there are many methods about the research methods of burst, such as mechanism analysis, field test, acoustic emission, energy theory, the strength theory, catastrophe theory, fractal theory. Due to the complexity of the phenomenon of rock burst, it is difficult to understand the mechanism in depth, therefore, it is difficult to establish prediction models with the traditional mathematics, mechanics. The current data of burst is not much, then how to extract features of audit data in the case of small samples is an important problem in order to to achieve better generalization ability This article introduces the field of machine learning to learn specialized in small sample statistical theory, using the most sophisticated statistical learning theory support vector machine method to solve such problems.With the computer technology and test methods improved, prospection based on field burst data is an important prediction method especially. Genetic Algorithm and Particle Swarm method combined to support vector machine not only give fully play to both the support vector machine good generalization ability, but also can play a global optimization algorithm of genetic algorithm and microparticles. The method is not only applicable to deal with small samples, nonlinear and other complex problems, but also applicable to avoid the difficulty of determining support vector machine parameters.Burst is divided into three risk levels:a serious impact on risk areas, medium-impact risk areas, no impact danger zone. Sphere multi-class support vector machine has a unique advantage to solve the problem of multiple classifiers and the article proposed a non-equilibrium situations in seeking support vector machine model. Numerical experiments show that this method has better prediction efficiency.