A hybrid rough sets and support vector regression approach to short-term electricity load forecasting

This paper aims to develop a load forecasting method for short-term load forecasting based on a hybrid approach, which combines the support vector regression method and the rough sets method. In the first stage, the rough sets method is applied to reduce the redundant attributes among varied factors that affect the short-term load forecasting. Then, a SVR module is trained using historical data reconstructed according to the attribution reduction results obtained by the first stage to perform the forecast. Numerical experiments on the historical data of Liaoning province grid in China show that, when compared against both neural network method and standard SVR method, the proposed method can forecast more accurate results while enhancing the training speed.