Abstract: As a clean and environment-friendly way of travel, electric vehicles are more and more popular, but the increasing charging load of electric vehicles will have a certain impact and impact on the existing power grid. Different from the conventional load, the charging load of electric vehicle has great randomness. In order to accurately predict the change of charging load of electric vehicle, firstly, K-means algorithm is used to cluster the charging load curve of electric vehicle at each station to reduce the fluctuation of charging load. At the same time, the charging load time series is a typical nonlinear and non-stationary time series. Therefore, multi-level wavelet change is introduced to decompose the charging load time series into multiple components with low complexity to help the prediction model mine its change characteristics. Then a long-term and short-term memory neural network prediction model is proposed, which takes all levels of components of historical charging load power, weather data and date type as inputs, and genetic algorithm is used to select the optimal super parameters of long-term and short-term memory neural network. Finally, the actual data verify that the proposed method can effectively predict the short-term load of electric vehicles.