基于机器学习的高压输电线路电晕损耗预测

This paper presents the application of machine learning in forecasting corona losses on high voltage overhead transmission lines. Forecasting of corona losses is challenging due to a highly-complex relationship between weather conditions and corona losses. To develop a forecasting model for corona losses, two years of weather data from different weather stations along the transmission lines and respective PMU measurement data from both ends of the line were considered. Machine learning algorithm-based forecasting models, e.g., XGBoost and ensemble random forest (ERF) regression models with multiple weather inputs, were used. The forecasting scenario was divided into four frames, i.e., different time-step based forecasting, seasonal forecasting, combined transmission lines corona losses forecasting for larger grid networks, and the effect of reduced features (weather parameters) on the accuracy of predicted corona losses. The best case performance was an error in the range of 0.5 kW/km for 98% of the data points and a root mean squared error of 0.16 kW/km. Precise forecasting of corona losses can assist Transmission System Operators in cost savings and increasing the reliability of the power system.