A Spatio-Temporal Model for Predicting Wind Speeds in Southern California

Abstract

The share of wind power in fuel mixes worldwide has increased considerably. The main ingredient when deriving wind power predictions are wind speed data; the closer to the wind farms, the better they forecast the power supply. The current paper proposes a hybrid model for predicting wind speeds at convenient locations. It is then applied to Southern California power price area. We build random fields with time series of gridded historical forecasts and actual wind speed observations. We estimate with ordinary kriging the spatial variability of the temporal parameters and derive predictions. The advantages of this work are twofold: (1) an accurate daily wind speed forecast at any location in the area and (2) a general method applicable to other markets.