| description abstract | The demand for renewable energy is first and foremost for environmental purposes and economic purposes. Wind energy, due to its many advantages, is receiving increasing attention. Thus, to evaluate wind energy properly, an accurate forecasting method is important for solving the problem. This paper proposes a hybrid model named discrete wavelet transform (DWT)—brain storm optimization (BSO)—back propagation neural network (BPNN) to forecast short-term wind speed, which uses a DWT to decompose the original wind speed into a low-frequency and high-frequency series, remove the high frequency noise sequence and reconstruct the remaining series. Next, the brain storm optimization algorithm is used to select the best parameters of a back propagation neural network. This paper also explores how to find the best parameter choice for the proposed model. To test the stability and availability of the developed model, a three-month wind speed time series from three observation sites of a wind farm located in Shandong Peninsula of China is used as the test set. Compared with linear regression, Auto Regressive Integrated Moving Average (ARIMA), BPNN, BSO-BPNN, Back Propagation Neural Network with Double Hidden Layer (BP-Hidden), First-order Adaptive Coefficient forecasting method (FAC), Elman Neural Network (Elman), Generalized Regression Neural Network (GRNN) and DWT-BPNN, the proposed model has better precision and robustness and is especially suitable for short-term wind speed forecasting at a large-scale wind farm in China. | |
