Short-term wind speed forecasting with ARIMA model

Abstract

The paper deals with the short-term forecasting of wind speed for the Laukžeme wind farm (Lithuania) using time series approach. The ARIMA model was selected and its best structure determined using the historical wind speed data (4 months) and varying both learning interval (3-5 days) of the model and the factual data averaging time (1-6 hours). The accuracy of forecasting was evaluated in terms of RMSE and absolute error. The forecasting results for 39 consecutive time intervals with 6-48 hourly forecasts are presented and discussed.