An innovative hybrid approach for multi-step ahead wind speed prediction

Abstract

Abstract It is essential to enhance the ability of wind speed forecasting for wind farm managers. The contribution of this research is to develop a novel hybrid method for multi-step ahead wind speed forecasting including empirical wavelet transformation-Kullback–Leibler divergence, autoregressive fractionally integrated moving average and improved back-propagation neural network. The empirical wavelet transformation-Kullback–Leibler divergence is used to extract these valuable features of wind speed fluctuations. The autoregressive fractionally integrated moving average is utilized to extract the long memory characteristics and capture the linear fluctuations of wind speed. The improved back propagation neural network is established to capture the corresponding nonlinear fluctuations, its inputs and outputs are determined by phase space reconstruction and its weights and thresholds are optimized by a modified bat algorithm with cognition strategy. Three prediction cases are employed to test the developed model. The simulation results demonstrate that the developed model outperforms several common benchmark models.