Short-Term Load Prediction of EV Charging Station Based on LSTM Recursion

With the rapidly increasing market share of electric vehicles (EV) in recent years, the randomness and uncertainty of EV charging load have posed challenges to the operation of distribution networks. To improve accuracy in load prediction of EV charging station, a short-term load prediction method based on the long short-term memory (LSTM) recursive neural network is proposed in the paper. First, the Canopy-based K-means clustering algorithm is improved to divide the similarly daily load sets of EV charging station. Then, the characteristic information of load days, such as meteorological data, is processed to obtain the typical feature vector of each load set. Next, the bidirectional long short-term memory (BiLSTM) models corresponding to each load set are trained based on the sliding window principle and phase space reconstruction. By calculating the Euclidean distance of the feature vectors between the predicted day and each load set, the most suitable pre-trained BiLSTM model is selected to predict the charging station load. Finally, simulation results demonstrate that the proposed method can effectively improve the load prediction accuracy.