To explore the hidden information in power load data and improve the accuracy of load forecasting, a hybrid prediction model based on Variational Mode Decomposition (VMD), Long-term and Short-term Memory Network (LSTM), Improved Particle Swarm Optimization (IPSO) algorithm and Gated Recurrent Unit Neural Network (GRU) is proposed to deal with the strong nonlinearity and non-stationarity of power load. First, use correlation analysis to determine the input factors, then use the VMD algorithm to decompose the load data into a series of intrinsic mode functions (IMF) and residual quantities combined with sample entropy decomposition to more reasonably determine the decomposition layer number and penalty factor. Secondly, according to the zero-crossing rate, these quantities are divided into low-frequency and high-frequency components. The low-frequency component uses the LSTM network, and the high-frequency component uses the IPSO-GRU network for prediction respectively. Finally, the predicted results are reconstructed to obtain the final result of power load.SThe simulation results show that compared with other conventional models, this hybrid model can effectively extract modal features and has higher prediction accuracy.