Short-term prediction of wind power generation based on VMD-GSWOA-LSTM model

Abstract

To improve the short-term wind power output prediction accuracy and overcome the model prediction instability problem, we propose a combined prediction model based on variational modal decomposition (VMD) combined with the improved whale algorithm (GSWOA) to optimize the long short-term memory network (LSTM) short-term wind power. First, VMD is utilized to decompose the wind power input sequence into modal components of different complexities, and the components are reconstructed into subcomponents with typical characteristics through approximate entropy, which reduces the computational scale of non-smooth sequence analysis. Second, the GSWOA is used to optimize the main influencing parameters of the LSTM model in order to obtain the weights and thresholds under the optimal LSTM model and to use the reconstructed individual subsequences. Finally, the actual data from two wind farms in Xinjiang and Northeast China are taken to verify the generalizability of the proposed model. The comparative analysis of the prediction results under different scenarios demonstrates that the improved model shows higher performance than the original model.