This paper explores how the accuracy of short-term prediction of wind speed and direction can be enhanced by considering the diurnal variation of the wind. The wind speed and direction are modelled as the magnitude and phase of a complex-valued time series. The prediction is performed by a multichannel filter using the spatio-temporal correlation between measurements at different geographical locations and the past values of the target site. A multichannel complex-valued non-stationary prediction Wiener filter is proposed that takes into account both the seasonal and diurnal variation of the wind. Using hourly wind speed and direction measurements from over 22 Met Office weather stations distributed across the UK, we demonstrate that there can be a benefit for predicting one hour ahead when taking into account the diurnal and seasonal cyclo-stationary nature of the wind.
|Title of host publication||International Work-Conference on Time Series|
|Number of pages||12|
|Publication status||Accepted/In press - 28 Apr 2015|
- multichannel predictive filter
- short-term wind forecast
Malvaldi, A., Dowell, J., Weiss, S., & Infield, D. (Accepted/In press). Short-term forecasting of wind speed and direction exploiting data non-stationarity. In International Work-Conference on Time Series (pp. 1-12). Springer-Verlag.