Explainability analysis of V2G aggregated capacity prediction

The integration of Vehicle-to-Grid (V2G) systems in smart grids requires accurate and interpretable predictive models to estimate Aggregated Available Capacity (AAC) and optimize energy flows. This study presents a proof of concept for explainable AI (XAI) methodologies applied to AAC prediction, leveraging a dataset of vehicle mobility patterns, meteorological conditions, and calendar effects in the metropolitan area of Rome. Both linear and nonlinear data-driven models were evaluated, considering FIR and nonlinear FIR using exogenous inputs, and ARX and nonlinear ARX considering autoregressive and exogenous input selections. Black-box nonlinear models were explained using Shapley Additive Explanations (SHAP). Results highlight the importance of exogenous variables in improving AAC prediction for the different intervals of the day. The findings emphasize the need for interpretable models in decision support systems for energy providers and V2G infrastructure managers, ensuring reliable and transparent decision-making.