Icing Forecast Of Power Lines Based On Support Vector Machine

In recent years, overhead transmission lines have suffered from icing accidents, which have affected the safe and stable operation of grid and caused serious blackouts. People's normal life has been affected and huge economic losses have taken place in power grid.Therefore, it is necessary to forecast the ice thickness of transmission line.At present, the ice thickness of transmission line, with the BP neural network prediction method, grey theory and RBF neural network, etc. The BP neural network slow convergence speed and easy convergence to the local minimal; Grey theory not exponentially with the original data, the prediction precision is low; RBF neural network to deal with high dimension problems will make learning more time-consuming. And support vector machine(SVM)algorithm can better deal with slow convergence speed and high dimension problems, so this article choose the support vector machine to forecast ice thickness of transmission line. The main work of this paper are summarized as follows:(1) Forming conditions, influencing factors, classification of transmission line icing and the mechanism of icing are made a deep analysis, And comparative analysis the advantages and disadvantages of several kinds of commonly used transmission lines ice cover model;(2) BP neural network, RBF neural network, gray theory and support vector machine(SVM) algorithm are used to simulate the ice thickness of transmission line. By comparison, although accuracy of RBF neural network model is better than BP neural network and gray theory, the penalty factor of support vector machine and kernel parameter values will directly affect the accuracy of the model, and the traditional support vector machine algorithm have not optimized and designed to penalty factor and kernel parameter values. In order to optimize the these two parameters, particle swarm optimization and genetic algorithm are introduced, and different penalty factor and kernel parameters are got by optimization. Based on the optimized parameters, the prediction model of ice thickness of transmission line is modeled and ice thickness of transmission line is predicted respectively.To evaluate the quality of the model, root mean square error indicators are introduced;(3) Design under the matlab GUI, provides different data transmission line ice thickness prediction software tools comparison, SVM algorithm that optimized by the genetic algorithm is superior to SVM algorithm that optimized by particle swarm, the former is more suitable for ice thickness forecast of transmission line.Research found that: 1) The ice thickness of transmission line, the data sample is limited,for this kind of situation, SVM algorithm is better than BP neural network; 2) The ice thickness of transmission line, the data sample containing multiple inputs and one output,belong to high dimension problems, when dealing with such problems, SVM algorithm is superior to RBF neural network; 3) In this data, support vector machine algorithm optimized by the genetic algorithm is superior to SVM optimized by particle swarm, the former is more suitable for ice thickness forecast of transmission line.