A Computational Implementation to Forecast Electric Vehicles Usage in the Power System

The mass adoption of electric vehicles (EVs) has been accelerated by the imperative need for energy transition. Nevertheless, the continuous growth of EVs can bring new challenges to electric power systems, including voltage and congestion constraints in low-voltage and medium-voltage networks. To anticipate these constraints, it is necessary to develop algorithms for predicting EV charging behavior. Therefore, this paper proposes a computational implementation based on Random Forest for accurately forecasting EV power consumption. The process begins with data collection, followed by pre-processing to clean and prepare the data. Next, feature engineering and feature selection steps are applied to create and select the inputs (features) for the forecasting process. Finally, forecast and validation steps are implemented. The case study analyzed utilizes real data from the charging station at Instituto Superior Técnico in Lisbon. The results demonstrate the effectiveness of the proposed forecasting method, achieving a Mean Absolute Error of 203.23W and a Normalized Root Mean Square Error of 3.44% for one year of training and one month of forecasting.