Power Cable Faults Recognition Based On The Support Vector Machine

Support vector machine (SVM) proposed in the last 90's century is a new statistic learning method based on fewer samples. Support vector machine, which based on statistical learning theory and structural risk minimization principle, is of the good extension ability and the better accuracy. It integrates the uses of optimal classification hyperplane, Mercer's theorem, convex quadratic programming and several technologies. It can solve effectively some problems such as over learning, dimension curse and local minimization point. Support vector machine, which is widely applied in much regions such as pattern recognition, regression estimation and so on, has became a research hotspot for its excellent learning performance in machine learning domain. Support vector machine is originally proposed for binary classification problems. How to extend it for multi-class classification is a significant issue.Aiming at the classification problem, the following contents are discussed from the prospect of the basic theory of the Support vector machine. (1) The machine learning, statistical learning theory and the development and research status of support vector machine are introduced. (2) For the two-class classification algorithm of the support vector machine, many kinds of support vector machine models existed at present are analyzed. (3) For the multi-class classification algorithms of the support vector machines, the advantages and the disadvantages and the performances are compared and summarized. The methods include the one versus rest, one versus one, binary tree, decision directed acyclic graph and error correcting output code. Furthermore, the convex hull binary tree support vector machine is developed based on the traditional binary tree to solve the problem of binary tree's structure, and the efficiency of improved methods are proved by the simulation experiment.The main works in this paper are about the structure detemining of the multi-class classifier of the support vector machines, the developing of the new algorithms of the multi-class classification of the support vector machine, the comparison relationship of the reparability ability between any two classes, and the connection with the kernel function. In addition, the method of the convex hull binary tree support vector machine is applied to the four state recognitions of the power cable, and a better results is obtained.