PR-KELM: Icing level prediction for transmission lines in smart grid

As a hot spot, the diffusion of the smart city has been more and more popular and widespread all over the world. The concept of smart city has supported a higher quality of urban spaces and a better offering of public services. Smart grid is an important component and the energy supply of smart city. The health of the smart grid will directly affect the health of the smart city. Efficient and accurate prediction about icing level of transmission lines provides a reliable basis for anti-icing and deicing of power grids. The prediction can not only significantly enhance the reliability, safety and stability of transmission networks, but also timely provide the state monitoring of ice on wires to the workers to make proper decisions and to prevent freezing disasters so as to effectively protect people's safety and build a stable and friendly society. To reduce catastrophic damages caused by the iced transmission lines of the grid, this paper proposed a novel hybrid model called PR-KELM, an integration of Principal Component Analysis, ReliefF and Kernel based ELM (Extreme Learning Machine) method to predict the icing level of transmission lines. This hybrid model covered two stages including: 1) structured and unstructured data feature extraction by local binary patterns, principal component analysis and ReliefF; 2) radial basis function Kernel based ELM predicting the processed data. The experimental data adopted in this paper were from the online monitoring system of South China State Grid Online Monitoring System by terminal CCO289 between December 1, 2011 and March 1, 2016. Experimental results indicated: compared with Elman neural network model, random forest model, support vector machine model and Bayesian networks model, PR-KELM approach demonstrated high accuracy and lower prediction error in terms of RMSE (Root Mean Square Error), MAE (Mean Absolute Error), MAPE (Mean Absolute Percentage Error), and R (Coefficient of Correlation) while applied to predicting in real datasets. (C) 2019 Elsevier B.V. All rights reserved.