Data-Driven Modeling of Electric Vehicle Charging Sessions Based on Machine Learning Techniques

Raymond O. Kene, Thomas O. Olwal

The increased demand for electricity is inevitable due to transport sector electrification. A major part of this demand is from electric vehicle (EV) charging on a large scale, which is now a growing concern for the grid power distribution system. The lack of insight into grid energy demand by EVs makes it difficult to manage these consumptions on a large scale. For any grid load management application to be effective in minimizing the impact of uncontrolled charging, there is a need to gain insight into EV energy demand. To address this issue, this study presents data-driven modeling of EV charging sessions based on machine learning (ML) techniques. The purpose of using ML as an approach is to provide insight for estimating future energy demand and minimizing the impact of EV charging on the grid. To achieve the aim of this study, firstly, we investigated the impact of large-scale charging of EVs on the grid. Based on this, we formulated an objective function, expressed as a sum of utility functions when EVs charge on the grid with constraints imposed on voltage levels and charging power. Secondly, we employed a graphical modeling approach to study the temporal distribution of EV energy consumption based on real-world datasets from EV charging sessions. Thirdly, using ML regression models, we predicted EV energy consumption using four different models of fine tree, linear regression, linear SVM (support vector machine), and neural network. We used 5-fold cross-validation to protect against overfitting and evaluated the performances of these models using regression analysis metrics. The results from our predictions showed better accuracy when compared with the results from the work of other authors.