Wind Speed Forecast Comparison Using WRF Model and Artificial Neural Network

Abstract The aim of this work is to improve wind speed forecasting, using the atmospheric mesoscale model Weather Research and Forecasting (WRF) and Artificial Neural Network (ANN) nonlinear auto regressive (with external input - NARX and without external input - NAR). The accuracy of the predictions was measured with observed data (OBS) measured every 10 min in an anemometric tower 50 m high, located in Craíbas (dry region of Alagoas State). The univariate statistics indicated that the forecasting represented well the wind temporal evolution in the studied period (April 2015). The average, maximum and minimum OBS speeds were 5.26 m.s−1, 12,29 m.s−1 and 0,01 m.s−1. Predictions ranged from 5.18 m.s−1 to 5.41 m.s−1 for the average, 11.58 m.s−1 to 13.92 m.s−1 for the maximum and 0.01 m.s−1 to 0.36 m.s−1 for the minimum. On the bivariate analysis, the statistical metrics used resulted in the following: Mean Error (ME) from -0.31 m.s−1 to 0.04 m.s−1; The root mean square error (RMSE) from 1.14 m.s−1 to 1.27 m.s−1; Mean Absolute Percentage Error (MAPE) from 22 to 23%; Correlation coefficient from 0.63 to 0.72. These results, despite considering a short period of data, indicate the potential for applying ANN and WRF in forecasting wind speed.