A hybrid hour-ahead wind power prediction model based on variational mode decomposition and bio-inspired LSTM

Abstract

Wind power, as an eco-friendly renewable energy, has been widely integrated into modern power systems. The prediction accuracy of wind power is crucial to the secure operation of power systems. To improve the prediction accuracy, a hybrid hour-ahead wind power prediction method is presented in this paper. First, the variational mode decomposition (VMD) is used to decompose the original wind power sequences into a set of intrinsic mode functions (IMFs) with different frequencies. Then, the prediction model is formulated by using the long short-term memory (LSTM) network for each IMF. To enhance the LSTM, a bio-inspired algorithm named Harris Hawks optimization (HHO) is blended to optimize the parameters of each LSTM prediction model. The final prediction output power is thus obtained by integrating all prediction results from these individual IMFs, so that the novel VMD–HHO–LSTM prediction strategy is developed. Finally, case studies based on the real historical dataset are performed, demonstrating that the proposed hybrid hour-ahead wind power prediction model named VMD–HHO–LSTM has better prediction performance and higher applicability to multiple dataset scenarios.