Operation of Ultra-Fast Charging Stations with Virtual Batteries on Voltage Constraints using Physics-Informed Neural Network

The proliferation of battery electric vehicles (BEVs) necessitates advanced ultra-fast charging station (UFCS) operation strategies to satisfy megawatt charging demands with the safety constraints of lithium-ion batteries. While existing virtual battery (VB) models facilitate aggregated charging control, they critically neglect dynamic voltage constraints and voltage variation limits inherent to battery dynamics, leading to overestimated power flexibility regions and operational risks. To mitigate this problem, we propose (1) an event-driven UFCS operation framework integrated with a physics-constrained VB model, where dynamic voltage and voltage variation constraints are rigorously derived from the battery RC model; (2) a physics-informed voltage estimation (PIVE) model that incorporates real-time battery information, including state-of-charge (SOC) and current, is developed to reconstruct voltage-SOC relationships into constraints of the UFCS operation framework. The PIVE, trained on a dataset of 10,000 BEV charging processes involving four types of widely used lithium-ion batteries, demonstrates superior voltage estimation accuracy compared to conventional data-driven models. In case studies, the violation frequency and SOC deviation severity indexes are developed to evaluate the influences of VB models on the BEV charging process. Experiments on the UFCS with 200 to 500 daily BEV arrivals demonstrate the effectiveness of dynamic voltage and voltage variation constraints in VB using the two indices.