Transfer Learning for Wind Speed Forecasting in the Desert

Wind energy and renewables, in general, would lead the way for changing the worldwide electricity industry and assist countries to meet Paris climate aims of reducing CO2 emissions by 2050. However, wind energy production is unstable due to weather changes. Managing smart energy systems mandates accurate power output prediction. Novel wind energy forecasting methods based on deep learning (DL) have achieved competitive performance compared to traditional methods. Nevertheless, they need big data for training and testing. This paper studies the feasibility of using transfer learning (TL) to develop DL models for hour-ahead wind speed (WS) prediction in a hot desert climate. It also presents the forecasting results of a DL model for data gathered from four locations in Saudi Arabia with different terrains to identify the success factors of using TL for WS prediction in a desert climate.