Comparison Analysis of Deep Learning Forecasting Models with Hybrid Structure for Short-Term Wind Power Prediction

Wind power penetration into the power system has been increasing in the recent decade. An accurate prediction of wind power is of great importance for energy dispatch, scheduling, and maintenance of the electric grid. This report conducts a comparative analysis of hybrid deep learning forecasting models for the short-term wind power prediction by combining powerful neural network modules such as convolutional neural network (CNN), bidirectional long short-term memory neural network (Bi-LSTM), and attention mechanism. In the proposed combinations, CNN is used to extract multi-dimension wind features through convolution and pooling operations, Bi-LSTM handles the sequential features fusion, and the attention mechanism computes the masks to extract critical features and predict the future wind power points. A comprehensive experiment for comparison between existing and proposed hybrid artificial neural networks is carried out to show the merits of the deep-learning hybrid models. Then, the effectiveness of the proposed model is verified with the historical data obtained from the National Renewable Energy Laboratory (NREL) website, and the results reveal that the hybrid combination can help fit the peak power output more accurately than the basic LSTM, GRU, and the auto-encoder networks.