Ultra-short-term Wind Power Forecasting Based on Deep Belief Network and Wavelet Denoising

Focusing on the limited generalization ability of traditional machine learning methods for wind power data, combined with deep learning theory, an ultra-short-term wind power prediction method based on wavelet denoising and deep belief network (DBN) is proposed. Firstly, the signal-to-noise ratio of wind power data under different wavelet basis functions is compared and analyzed. And then selected db4 wavelet function to decompose the historical data in three layers. The wavelet soft threshold method is used to remove the high- frequency noise signal in the data, which is sent from the power plant to the master station of the power dispatch system. Based on the historical wind power data after de-noising, a DBN based wind power prediction model is constructed, which is optimized by unsupervised pre-training and BP algorithm. The DBN prediction model can effectively extract the characteristics of wind power data, and has good prediction performance and data generalization ability. Experiments with actual power system data, the results show that the proposed wind power prediction method has a better prediction effect than traditional machine learning algorithms such as BP and RBF.