LSTM and Linearized Segmented Model for Short-Term Power Output Forecasting of Wind Turbines

Abstract

In this paper, an indirect model for predicting wind turbine output power is proposed, based on a LSTM and a mathematical model for wind power estimation. The LSTM model is used to forecast future changes in wind speed and direction, and the linearized segmented model to estimate wind turbine output power based on forecast wind speeds. To train the LSTM model, NASA meteorological data were used for a period from January 1,2015 to December 31, 2022 for a coastal site in Benin. The results obtained showed that the forecasts made by the LSTM model are fairly close to the actual values. Values of 0.36 m/s and 8.1° for the Mean Square Error (RMSE) and 0.92 and 0.71 for the Coefficient of Determination (R2) were obtained for wind speed and direction forecasts respectively. The predicted speeds were then used to estimate the power output of the NE-200S wind turbine.