Improvements in wind power prediction from locational wind speed estimates: a critical analysis

Abstract As the energy transition process is accelerating, firmer and more accurate power supply and demand forecasts become paramount. A lot of effort is put by the transmission system operators or market participants to quantify the amount of renewable power. Due to intermittence and data dissemination, forecasting wind power generation at system level remains a challenge. The current research aims at highlighting the benefit of employing locational wind speed estimates in predicting grid-wide wind generation. We use ordinary kriging to downscale a spatio-temporal grid of wind speed data to the wind farm locations. Then, we build power curves based on Betz law that we eventually aggregate and optimize to account for the whole power generation of Southern California. We quantify the model performance on the generic grid points versus on the actual wind farm locations. We show that the reduction of in-sample forecasting error is of 3.6% (in root-mean-square error) and none on the out-of-sample domain. Additionally, we extend our model by employing a novel kriging application on model parameters. We show that the improvements from locational wind speeds are limited in this case. Finally, we raise several critical points about these approaches.