Meteorological processes information mining and transmission lines icing forecast

In view of the imperfection of existing predictive regression models for icing forecast where only meteorological parameters at one moment are used to foretell the predicted icing value at the same moment, a revised icing forecast method based on meteorological process information mining is proposed. The samples of meteorological parameters are divided into three fuzzy pattern categories, i.e., icing growing, sustaining and melting. Then the membership function with the variable symbolized by Mahalanobis distance from meteorological parameters sample to the center of categories is defined. And the method of gray slope-correlation is used to determine the weights of meteorological parameters to evaluate the Mahalanobis distance. The degrees of membership and the samples of meteorological parameters are combined to form the high dimensional historical samples containing icing meteorological processes information, then support vector machine method is used to train the icing forecast regression function and predict. Numerical tests are conducted and the results indicate that the mean relative error in neural network method gets 24.50% and 22.66% in the current support vector machine method, but 6.62% in the revised cases. The icing forecasting model based on the meteorological processes information mining is endowed with higher predicting accuracy.