Modeling Uncertainty and Predicting Electric Vehicles Charging Behavior Through Stochastic Poisson Processes

Abstract

Electric vehicle charging behavior is influenced by various uncertain factors, presenting challenges for accurate modeling and prediction. This study proposes a straightforward approach employing stochastic compound Poisson processes to represent daily electric vehicle supply equipment charging patterns, which takes into account the number of hourly events, the average charging power and the duration. After a review of the available Poisson process techniques, the methodology provides a parameter evaluation and a validation procedure based on the comparison of real data and model outcomes such as the daily energy consumed and the daily maximum power absorbed from the electrical network. The analysis, using real-world electric vehicle charging stations data, demonstrates the ability of the homogeneous compound Poisson process and of the double stochastic compound Poisson process to characterize charging events as stochastic processes. It is then possible to predict daily electric vehicle charging stations average power and also to evaluate forecast accuracy with respect to the amount of data used to estimate models parameters. It is shown that using stochastic Poisson processes allows to have good prediction models for electric vehicle charging stations even with a small historical dataset available, and hence could be used to support the ongoing mobility transformation.

Keywords: stochastic modelling, electric vehicle load modelling, compound Poisson processes, energy forecast, grid digitalisation