Short term demand forecasting of electric vehicle charging stations using context aware temporal transformer model

The growing number of electric vehicles (EVs) on the road poses great challenges to the power supply and causes outages. Most existing research works focus on individual or aggregated charging station data at the city level. However, charging behaviors at different city locations might demonstrate different patterns and characteristics. This study proposed a Context-Aware Temporal Transformer (CAT-Former) model using Temporal and Contextual features for short-term EV charging demand forecasting of one hour and one day ahead using the public EV data from Boulder City, Colorado. The temporal and contextual features are important features, which help the model to understand the charging patterns of different periods over different locations. The charging data with different trends is crucial to train and test the proposed model performance. Therefore, this study chose the three locations with the highest number of sessions from the data. The performance of the proposed model, as well as the baseline models, including LSTM, BiLSTM, and hybrid models such as CNN-LSTM and CNN-BiLSTM, is assessed and compared using Mean Square Error (MSE) and Mean Absolute Error (MAE) on three locations. The proposed model is compared to the Simple and Hybrid Transformer models utilizing the LSTM-based Encoder-Decoder. The proposed model performed better than the baseline models for one hour and one day ahead of forecasting for the selected locations by achieving the lowest MSE and MAE values. The results show that the proposed CAT-Former model using temporal and contextual features can effectively forecast the charging demand using charging data from different locations for short-term periods, including one-hour and one-day ahead predictions.