Wind Power Forecasting using Generalized Autoregressive Score Model

Abstract

Accurate Wind Power Forecasting (WPF) is essential for the economic operation and planning of electric power systems and electricity markets. Wind power forecasts up to few hours ahead (very short-term) are utilized for optimal operation of power systems. Several time-series models are proposed in the existing literature for Very Short-Term Forecasting (VSTF) of wind power. These include Autoregressive Integrated Moving Average (ARIMA), Generalized Autoregressive Conditional Heteroskedasticity (GARCH), Hybrid ARIMA-GARCH, etc. Although these models are mathematically well off and effective for short-term WPF but not accurate because of their fixed or time-independent parameters. Therefore, this paper presents a novel Generalized Autoregressive Score (GAS) model for WPF considering time-varying parameters. GAS model parameters are updated online for each forecasting lead time by a feedback system. The proposed model is implemented on three Australia-based wind farms and obtained results are compared to benchmark ARIMA and ARIMA-GARCH Hybrid models. The simulated results show that the GAS model has the highest accuracy and offers minimum error followed by ARIMA-GARCH Hybrid, and then ARIMA.