Power load is characterized by nonlinearity, non-stationarity, and strong variability. Traditional time series forecasting methods are not sufficient to capture its complex features and evolving patterns. To improve the accuracy of short-term power load forecasting, an improved VMD-EEMD-LSTM hybrid model was proposed. First, historical power load data were decomposed by variational mode decomposition (VMD), and intrinsic mode components and a residual term at different frequencies were obtained. Then, the residual term generated by VMD was further decomposed by ensemble empirical mode decomposition (EEMD). After decomposition, the intrinsic mode functions (IMFs) obtained from VMD and EEMD were used as inputs to a long short-term memory (LSTM) network for forecasting. The final forecasting result was obtained by combining the predicted values of all components. Comparative experiments were carried out using actual power load data from a region in Jiangxi Province. The results show that, compared with the VMD-LSTM model, the root mean square error (RMSE) and mean absolute error (MAE) of the proposed model are reduced by 56.57% and 56.03%, respectively. Compared with the EEMD-LSTM model, the RMSE and MAE are reduced by 10.06% and 9.01%, respectively. The results indicate that the proposed model achieves higher forecasting accuracy.