Very Short term Wind Power Prediction Using Hybrid Univariate ARIMA-GARCH Model

The integration of wind power generation with the grid reflects many challenges to the utility and market operators. For this purpose, very short-term Wind Power Prediction (WPP) has become an indispensable requirement for efficient power systems operations. Typically, statistical time series models like Autoregressive Integrated Moving Average (ARIMA) is widely used for WPP. Although ARIMA is capable of capturing conditional mean appropriately. However, it does not cover the time-varying volatility present in wind power. Hence, for more precise parameter estimation and forecasting, this paper utilizes the Generalized Autoregressive Conditional Heteroskedasticity (GARCH) model. To fortify the forecasting model, this paper proposes a Hybrid ARIMA-GARCH model, incorporating the assets of ARIMA and GARCH model, for forecasting of wind power. The proposed model combines the ARIMA based conditional mean forecast and GARCH based conditional variance forecast. The propounded model is implemented on three wind farms located in Australia. The simulation results obtained shows that the proposed model works better than conventional ARIMA Model in term of Mean Absolute Error (MAE) and Root Mean Square Error (RMSE).