Research on Electric Vehicle Charging Load Forecasting Based on the MCBDL-TCN Model

This study introduces a new method for predicting the charging load of electric vehicles (EVs) by combining a Temporal Convolutional Network (TCN) and Multiple Cascaded Bayesian Deep Learning (MCBDL) model. The study focuses on short-term EV charging demand prediction in an urban region of a northern Chinese metropolis, utilizing data from 38 public DC charging stations. Long-term trends and short-term oscillations in load data are both well captured by the MCBDL-TCN model. Experiments show that MCBDL-TCN is superior to LSTM and ANN traditional models in terms of RMSE, MAE and MAPE indexes. The results of predicting the four seasons, as well as workdays and non-workdays, demonstrate the model's outstanding performance in forecasting and addressing uncertainty, thus enhancing the accuracy and flexibility of complex time series data. This shows that the ability of the model to deal with uncertainty and extract rich feature information from long time series data is improved by combining LSTM-BDL unit with TCN.