Abstract:  In order to improve the short-term load forecasting efficiency and precision of electric vehicle charging station, this paper proposes a short-term load forecasting method for charging station based on kernel principal component analysis (KPCA) and non-dominated sorting genetic algorithm II (NSGAII). The KPCA is used to reduce the noise of the model input variables, which simplifies the network structure and accelerates the prediction speed. Through the comparison of the load forecasting error to define the convolutional neural network (CNN) model in convolution layers and sub sampling the top layer neurons number, the accuracy of the model is guaranteed. By using the NSGAII to optimize the parameters of the CNN, the operation speed and precision of the prediction method are improved. Through example analysis and comparison with other methods, it is proved that the method has high efficiency and precision.

Key words:  electric vehicle charging station, short-term load forecasting, kernel principal component analysis (KPCA), non-dominated sorting genetic algorithm II (NSGAII), convolutional neural network (CNN)