Bayesian Hierarchical Model for Characterizing Electric Vehicle Charging Flexibility

The proliferation of electric vehicle (EV) adoption introduces a suite of opportunities and risks to power system operation which have the potential to significantly increase loading, reduce equipment lifespan, impact rate design and offer energy flexibility. The uncertainty in space, time, power and energy presented by EV charging demand creates a unique challenge to the development of utility best practices. Specifically, this work seeks to predict the aggregate number of EVs initiating charging sessions per hour within a region's EV charger fleet. Therefore, we propose a Bayesian hierarchical model to learn the EV arrival as a stochastic process. The model is trained and tested by a 20/80 data split. Sampling convergence is confirmed by observation of the Gelman-Rubin statistic and predictions are made against the test data. Results indicate a prediction error of less than 3% over the testing data set.