Abstract: The strong randomness and high volatility of electric vehicle（EV） charging behavior make the short-term charging load prediction accuracy of its charging station low, and as a mobile power storage and load resource participating in the vehicle-to-grid（V2G） service, its scheduling centre needs to predict the charging load of the EV in a short period of time in order to improve its impact on the power grid load.In order to improve the accuracy of short-term charging load prediction for EV charging stations, a Crested Porcupine Optimizer-Variational Mode Decomposition-Bidirectional Long Short-Term Memory,（CPO-VMD-BiLSTM） combination model for short-term charging load prediction of EV charging stations was proposed.Firstly, multiple factors affecting EV charging loads and historical charging loads of charging stations are considered to form the input feature matrix.Then the CPO algorithm is used to optimize and search the core parameters of VMD to achieve the adaptive optimization settings of parameters.After that, CPO-VMD is employed to decompose the historical charging load data, weaken the non-stationarity of the load and capture its local characteristics.Finally, the decomposed feature matrix is input into the BiLSTM model to achieve the prediction goal of short-term charging load of charging stations.Using the historical charging load data of EV charging stations in the open data set of American ANN-DATA,which is located in the campus of California Institute of Technology as a practical example, compared with independent models, non-optimized combination models, and optimized combination models, the Root Mean Squared Error（RMSE） and Mean Absolute Error（MAE） are reduced by 41.23% and 59.04% on average.Therefore, the accuracy improvement and practicality of the proposed method in short-term prediction of charging load in charging stations are verified.