Short-term Charging Load Forecasting for Electric Vehicles with a Multilayer LSTM Model Considering Sliding Windows and Online Learning

As the large-scale development of electric vehicles (EVs), accurate short-term charging load forecasting for EVs is the basis of vehicle to grid (V2G) interaction. In this paper, considering the uncertainties of EV users' charging behavior, a multi-layer Long Short-Term Memory (LSTM) model considering sliding windows and online learning is proposed for short-term forecasting. Several typical features including peak and valley tariffs, environment temperature, charging date, charging time, historical average charging load are extracted considering their correlations with charging load. A multilayer LSTM model considering sliding window is proposed by inputting the typical feature matrix, and an online learning method is added to the model to improve the processing speed of the model for fast and accurate short-term charging load prediction. Finally, the proposed multilayer LSTM model considering sliding window and online learning is demonstrated to have unique advantages in terms of robustness, processing speed and prediction accuracy through practical examples.