A Hybrid CNN-LSTM-Attention Deep Learning Model for Forecasting Time-Series Electric Vehicles Fast Charging Loads at Public Stations

With large-scale adoption of Electric Vehicles (EVs), forecasting of charging loads is becoming crucial for utilities and network operators to understand the impact, plan for infrastructure upgrades, and optimally manage the operation. Uncertainty in EVs public charging along with fast ramping charging profile in fast chargers add more challenges to the forecasting process and make the conventional methods ineffective. This study proposes a new methodology for forecasting EV charging loads at public fast charging stations considering time-series data. The methodology is based on a combined Convolutional Neural Network (CNN) with Attention Long ShortTerm Memory (LSTM-Attention) algorithm that is designed to capture the stochastic nature of EV driver behaviors through both temporal dynamics and nonlinear time-series characteristics in the dataset. Real-world data from a fast-charging station in Pasadena, CA, along with weather data from the National Renewable Energy Laboratory (NREL) are considered to test and validate the proposed approach. The performance of the hybrid CNN-LSTM-Attention technique was compared with other forecasting techniques including Convolutional Neural Network (CNN), Recurrent Neural Network (RNN), and Long ShortTerm Memory (LSTM). The results indicate that the hybrid model, combining CNN and LSTM-Attention, outperforms other methods showing the least Root Mean Square Error and Mean Absolute Error, and the highest $R^{2}$, compared to other techniques.