Electric Vehicle Charging Station Load Prediction Based on Time Series Decomposition Provisionally Accepted

Yuheng Cai^1* Mingxuan Li¹ Yufan Cheng¹

• ¹Southeast University, China

To better capture the temporal dynamics of electric vehicle(EV) charging station loads and enhance the accuracy of load forecasting, a novel multi-step short and medium-term prediction framework based on time series decomposition is proposed. The volatility of time series and its computational complexity are common challenges affecting forecasting tasks. Addressing the volatility of charging station loads, a hierarchical decomposition framework is introduced to extract regular temporal information at each layer. Subsequently, predictions are made on the decomposed input sequences, and the final output is synthesized. To reduce computational complexity, linear and convolution operations are employed to construct the prediction model, incorporating an Average Pooling(AvgPool) layer as the initial decomposition sub-model. Finally, simulation validation is conducted using real charging load data from EV charging stations in Changzhou, China. The results indicate that the proposed prediction model exhibits a 27.8% decrease in mean square error (MSE) compared to the traditional LSTM model and a 10.4% decrease compared to the latest time series decomposition model, N-HiTS. This validates the superiority of the designed charging station load forecasting approach.