Load Forecasting for Electric Vehicle Charging Stations Using Sequence Decomposition and Deep Learning Techniques

The penetration of electric vehicle charging stations (EVCSs) is increasing rapidly with the development of large-scale transportation electrification. Due to the high randomness of user charging behavior, the charging load collected by the smart energy measurement system exhibits significant volatility and strong uncertainty, making it difficult to accurately predict the EVCS load. An EVCS load forecasting method is proposed by integrating sequence decomposition and deep learning techniques. First, the interpretable periodic and volatile load features of the multivariate load sequences are extracted by the multivariate singular spectrum analysis (MSSA) method to improve the robustness of the forecasting model. Second, the unique architectures of the bidirectional long short-term memory (BiLSTM) and the patch time-series Transformer (PatchTST) models are applicable to both interpretable load features, naturally improving prediction accuracy and robustness of the proposed method. Finally, simulations are conducted on a real-world dataset of EVCSs in a city center in China. The simulation results and comparative analysis indicate that the EVCS load forecasting method significantly improves prediction accuracy and robustness.