Staged icing forecasting of power transmission lines based on icing cycle and improved extreme learning machine

With the frequent occurrence of extreme weather, power transmission line icing, a general phenomenon during winter which may bring serious economic losses to the electric power system, has attracted increasing attention. However, owing to the complexity nature of wire ice covering, it is essential to establish an icing thickness forecasting model with high-accuracy for guaranteeing the security and stability of the power grid. Hence, this paper proposes a hybrid model that combines wavelet transform (WT) with extreme learning machine optimized by bat algorithm (BA-ELM). The original icing data containing icing thickness and meteorological factors are first denoised by WT and then divided into several stages based on the characteristics of icing period. Bivariate correlation analysis and partial auto correlation function (PACF) are used to select the inputs of different stages. Subsequently, ELM whose input weights and bias threshold were optimized BA is built to forecast icing thickness. To verify the developed model, icing data from two power transmission lines located in Hunan province are applied for experiments. The simulation results demonstrate that not only the proposed model shows a better performance but also the staged modeling can highly improve icing thickness prediction accuracy.