Keywords: wind energy conversion system, WECS, Elman recurrent neural networks, ERNN, mean absolute error, MAE, standard deviation of absolute error, Std, wind power, energy generation
Application of neural network to wind energy conversion systems
In the previous study, the energy generation of two WECS deployed at Darling site were estimated using the statistical–based technique. A total energy generation of 178.43 MWh at 50 m and 196.17 MWh at 70 m height were estimated from both WECS for the month of January 2010. In this study, the energy generations of both WECS for the same month were estimated using a developed wind power predictor. The recorded five–minute averaged weather data for the period of 32 days (31st December 2009 to 31 January 2010) were obtained for this study. The long term power generations of both WECS in time steps of five–minute of up to one–week ahead are predicted for the month of January 2010. The forecast results are compared to the actual energy generations obtained from the statistical–based technique. The forecast model returns an absolute error value of 5.265%, mean absolute error value of 0.402%, and standard deviation of absolute error of 4.504% for the 1.3 MW WECS power prediction. For the 1.0 MW WECS, an absolute error of 5.860%, mean absolute error of 0.419%, and standard deviation of absolute error of 5.398% were estimated. Furthermore, the usable energy generations of the WECS for the month are estimated.