Short-term generation forecast of wind farm using SVM-GARCH approach

Abstract

Wind generation forecast is important for power system operation, trading, and some other applications. In this paper, a practical approach for short-term wind generation forecast is proposed. The proposed approach uses Support Vector Machine (SVM) to produce the primary wind farm generation forecast results. However, since the residual error is assumed to be independently identically distributed (IID) in SVM, which ignores the strong volatility property of wind generation, a Generalized Autoregressive Conditional Heteroscedasticity (GARCH) model, which can predict the varying residual error, is used here to correct the SVM forecast results. The proposed approach can provide more reliable forecast results comparing with the usual SVM approach. Test results on two wind farms located in Heilongjiang Province in northeast China demonstrate the effectiveness of the proposed approach.