Data-driven Method Of Transmission Lines Icing Load Forecasting Model Studies

The disasters with icing of overhead transmission lines seriously affect the stability and safe of the power transmission system, it is important and meaningful for the subject of the study with icing prediction. Due to the complexity of the icing process on transmission, leading to the problem of uncertainty and multi-granularity with prediction model, it is difficult obtain the analytical model based on mechanism. On the other hand, the monitoring device on transmission collected a large number of micro-meteorological and icing data, so this paper used the study of icing prediction model based on data-driven, the main research achievement as follows:First the paper introduces the latest results of model studies with icing on transmission, pointed out the problems of icing forecast model, and analyzed the main factors with affecting the icing process from the perspective of the icing formation mechanism. As a result, the icing on overhead transmission lines mainly depend on the micro-meteorological factors.Second this paper constructs the polynomial fitting and moving average and autoregressive model from a linear perspective. This type model used the micro-meteorological as input, and adjusted the parameters of the model by training the historical data, it achieve the short-term prediction of icing on transmission, at last it also analyze and compare the advantages and disadvantages of the two models in the application of icing on the transmission.Last this paper constructs the prediction model from the nonlinear perspective, and reconstructs the phase space of micro-meteorological factors and icing load based on the chaotic time series. On this basis, it is proposed the machine learning methods icing load forecasting model based on BP neural network and support vector machine, and verified the effect of prediction model of the two types in the same transmission line on the same and different icing process from application. According to the results, compared with BP neural network, the SVM model has the advantage of high training efficiency and generalization ability, model stability and so on, it is better to predict the trends during the long-term and short-term on transmission lines, this model offer practice guidance measures on the maintain and melt the icing on the transmission and ensure the safe and stable of the power system.