Joint Occupancy and Load Profile Prediction for Electric Vehicle Charging Stations

Predictions of occupancy and power demand in charging stations for electric vehicles are challenging problems of the greatest relevance nowadays, as electric vehicles are becoming popular green transportation solutions. These forecasts are essential for scheduling charging profiles that optimally exploit renewable energy production. In this paper, we propose a methodology to predict the electric load and occupancy state of electric vehicle chargers in a synchronous manner, using deep learning frameworks. In particular, decision tree-based predictors have been trained using publicly available databases of actual charging stations. We derive mixed occupancy-load predictors for individual chargers, with a sampling time of 15 minutes, and a prediction horizon of 24 hours. The best prediction models yield an average occupancy state classification accuracy of 81%, mean absolute error (MAE) of 4.48 A, and root mean square error (RMSE) of 7.6 A on validation data.