Study On Prediction Of Transmission Line Icing

Overhead transmission line is a vital component of the power grid transmission of electric energy. Thus it’s necessary to ensure the safe and stable operation of transmission lines. However, in recent years there have been many ice disaster accidents of overhead transmission line at home and abroad, bringing harm to the safe operation of power grid. Severely, they can cause blackout accidents in large areas and make the grid difficult to recover. Prediction of the icing condition of the electric transmission line can let us timely know the trend of the ice on the wires and give warning. In this way, we can effectively prevent some serious accidents. Meanwhile, the prediction provides a basis for anti-icing and deicing on transmission lines.This paper first expounds the conditions to produce ice on transmission lines, mechanism and process of icing, its classification, and various influential factors and explains commonly used measures of anti-icing and deicing. Then the paper predicts icing using the influence factors of transmission lines and uses historical data of the influence factors as the input of the prediction model. In the experiment, the author first uses RBF neural network and support vector machine (SVM) model to forecast the icing, and gets the result that the prediction accuracy of SVM is higher than that of RBF neural network. It is known that penalty parameters and nuclear parameter of SVM have a great influence on precision of algorithm, so we can use genetic algorithm (GA) and particle swarm optimization (PSO) to optimize the penalty parameters and nuclear parameter. Besides, GA can be optimized by crossover probability and mutation probability, and PSO can be optimized by inertia factor and accelerated factor. Therefore, the author first improves the parameters of the two algorithms, and uses the improved algorithms to optimize penalty parameters and nuclear parameter in SVM and then the optimized SVM is used to forecast icing on transmission lines. The results show that in general, prediction time of the optimized SVM is increased, and so it is with the prediction accuracy. Through analyzing the data, the author finds that the SVM optimized by improved GA takes the shortest time to forecast, while the SVM optimized by the combination of improved GA and PSO the longest. What’s more, the SVM optimized by improved GA shares approximately the same prediction accuracy with the SVM optimized by improved PSO, while the SVM optimized by the combination of improved GA and PSO owns the highest prediction accuracy.From the experiment, we can learn that SVM is more suitable for dealing with small samples, for it predicts better than neural network when the historical data samples is lesser. And the experiment succeeds in improving the accuracy of the prediction of ice thickness by using the optimized measures to optimize SVM to increase the complexity of the prediction algorithm.