EV charging stations load forecasting by BiLSTM and PatchTST

The simultaneous integration of large-scale, high-power load of electric vehicle charging stations (EVCSs) significantly impacts the safe and stable operation of the distribution network. The stochastic nature of user charging behaviours poses significant challenges to traditional load forecasting methods. A load forecasting method for EVCSs based on deep learning techniques is proposed to improve the forecasting accuracy. First, a moving average approach is used to decompose original load sequences into trend and seasonal components. Then, the bidirectional units of the bidirectional long short-term memory (BiLSTM) model are employed to capture trend load patterns. The patch-based segmentation strategy and the attention mechanism of the patch time series Transformer (PatchTST) model are applied for the first time to capture seasonal load patterns. Finally, the load forecasting outputs from both models are summed and reconstructed to obtain the load forecasting results. The simulation results on a real-world dataset in China demonstrate the effectiveness and outperformance of the proposed method.