Model Identification Frameworks for V2G Available Aggregated Capacity Prediction: Transferability and Explainability

This study proposes a predictive modeling framework for Aggregated Available Capacity (AAC) estimation in Vehicle-to-Grid (V2G) systems, addressing the research gap in the integration of accessible real-world data, transferability and explainability of models. By using floating car data and incorporating exogenous inputs such as meteorological, calendar and traffic data, the approach improves prediction accuracy and model generalization. Data-driven linear regression and state-space models as well as nonlinear models are identified and compared. Transfer learning enables model adaptation across different V2G hubs, reducing the need for retraining and improving scalability across different urban environments. Explainability techniques such as SHAP provide transparency and promote confidence in predictions by providing decision makers with insights into key predictive factors. This combination of transferability and explainability strengthens the applicability of the framework to real-world V2G operations and enables more reliable and data-efficient predictions. The results support improved AAC predictions and help grid operators to optimize V2G services.