Wind Speed Forecasting by Box-Jenkins Models

The possibility of modelling observed wind speed time series and forecasting their future values is presented in this paper. Seasonal autoregressive integrated moving average (SARIMA) models are applied to time series formed by four years hourly average wind speed measurements in thirty sites of Sicily. Our approach is considerably different from the original one (the Box-Jenkins approach) since it is completely automatic. We use a peculiar feature of wind speed on a land area, its daily period, to identify a class of SARIMA models within which to find the best fitting model by information criteria (here we employ AICC). Here we report the results, concerning the fit and forecast accuracy, for five sites, chosen because of their very different geographical and orographical characteristics. The results obtained, through an appropriate time series stratification and difference, show that an automatic use of Box-Jenkins models, for wind speed time series modelling and forecasting on land areas, is worth considering