With the fast progress of electric vehicles, the load of charging stations plays an increasingly important impact on the power grid. To guarantee the safe operation of power grid, a prediction model with the Variational Mode Decomposition (VMD) and Long Short-Term Memory (LSTM) is investigated. Firstly, considering that there exist strong randomness and volatility in the data of electric vehicles charging load, we adopt the VMD algorithm to decompose the data into three modal components, for reducing the complexity of original data. In addition, decomposition error is also been decomposed into two modal components, for extracting the hidden information in the error. Thus, five subsequences are respectively predicted by the LSTM method, and summation the prediction results constitutes the forecasting of the charging load. Furthermore, experimental analysis indicates that the model is more appropriate for predicting the charging load.