Forecasting public use of EV charging stations is challenging as it can depend on time of day, price, EV driver behaviour, and the models for charging stations that currently exist tend to predict individual variables. The framework pro posed in this work is able to predict simultaneously the energy demand, the charging price, the charging stations occupancy, and the session duration using 4,344 real-world charging sessions over 6 months with 273 charging stations.The RF + XGB stacking ensemble achieved the best average R2 of 0.869 and MAPE of 10.21%, versus 0.722 and 15.42% for linear regression.

EV charging, ensemble learning, forecasting feature, CNN LSTM hybrid model, feature importance analysis, dynamic pricing prediction