Abstract: Improving the accuracy of load forecasting is very important to guide the production planning, economic dispatch and stable operation of power system. A short-term load forecasting model based on variational mode decomposition（VMD） and long short term memory（LSTM） neural network is proposed in this paper. VMD algorithm is used to decompose load series into different intrinsic mode functions（IMF）, and each IMF is combined with LSTM network for prediction. The prediction results of VMD-LSTM model are obtained by superimposing the prediction results of each part. The analysis of experimental simulation shows that compared with the single LSTM and empirical mode decomposition（EMD） combined LSTM prediction method, the proposed method can effectively improve the prediction accuracy.