Multi step ahead forecasting of wind power by different class of neural networks

This paper presents a comparison of three different classes of artificial neural networks (ANN) for multi-step ahead time series forecasting of wind power. The neural network needs past wind generation measurement as an input. For time series prediction, the time lag data pattern is required & for this purpose the statistical tool called autocorrelation function (ACF) facilitates to work out on the input variables of neural networks. The three models which have been used are: linear neural network with time delay (LNNTD), feed forward neural network (FFNN) and Elman recurrent neural network (ERNN). The performance comparisons of the models are on the basis of mean absolute error (MAE) & mean absolute percentage error (MAPE). Data of wind power from Ontario Electricity Market for the year 2011-2012 has been considered for the case study and tested for a period of one week for twelve multi-steps ahead forecasting. It is observed that all class of neural networks shows almost equal results.