A hybrid deep learning model based on parallel architecture TCN-LSTM with Savitzky-Golay filter for wind power prediction

Wind energy is experiencing rapid global growth. However, wind power generation time series data often exhibit nonlinear and non-stationary characteristics, which make precise estimation challenging. Consequently, wind power prediction assumes an increasingly vital role in the planning and deployment of power and energy systems. Recently, many hybrid deep learning prediction models have been developed to improve the prediction performance of wind power, but their deeper network layer and complex structure also bring higher computing costs and reduced prediction efficiency. In order to achieve higher prediction performance, reduce the complexity and computational cost of hybrid deep learning models, and improve prediction efficiency, this study proposed a hybrid deep learning model based on parallel architecture by using a tensor concatenate module to combine a temporal convolution network (TCN) and a long short-term memory (LSTM) neural network for wind power prediction, and the Savitzky-Golay (SG) filter is used to remove noise and smooth the input wind speed time series in the model training stage. Using a wind turbine case from Turkey, three sets of comparison experiments are conducted. The effectiveness and superiority of the proposed model are validated by comparing a variety of single and hybrid models using current evaluation metrics and the Diebold-Mariano test. Among them, the number of training parameters and computing time of the proposed parallel architecture TCN-LSTM hybrid model are reduced by 6.59% and 25.82%, respectively, when compared to the conventional TCN-LSTM hybrid model with the same hyperparameter settings. nMAE, nMSE, and nRMSE are reduced by 2.00%, 9.21%, and 4.74%, respectively. The Diebold-Mariano test results also reveal that the proposed model performed better in terms of prediction performance. Moreover, the proposed innovative architecture hybrid model provides a novel approach to developing a hybrid model of deep learning networks for wind power prediction.