Comparison of the wind nowcasting generated by the WRF model and an two LSTM models

Abstract

Cuba is immersed in the use of wind energy. However, for its development it has required various efforts in different fields, including the improvement of tools that make the wind predictable and, in turn, wind generation, such is the case of very short-term forecasts. For this reason, this paper compares the wind forecast of the Weather Research and Forecasting model (WRF) at 3 km spatial resolution a Long Short-Term Memory (LSTM) model type. The comparison and evaluation of the forecasts of the models is carried out with data from the Gibara I and II wind farms and the Los Cocos wind survey mast, located in Holguín, Cuba, with wind speed measurements every 10 minutes at a height of 50 m. The LSTM were built by first training the observations and then combining the observations with the WRF model forecast. The results of the comparison were carried out for three study cases and indicate that both LSTM models present better results than the WRF model, although the differences do not exceed 1 m/s. However, for the case studies, the WRF model behaves well reproducing the daytime cycle, but with a MAE greater than 4 m/s.

Keywords

• wind nowcasting
• renewable energy sources
• LSTM models
• WRF model