Learning-Assisted Aggregate Flexibility Characterization for EV Charging Stations Considering Scheduling Deviation Uncertainty

As electric vehicle (EV) integration scales, EV charging stations have emerged as valuable assets on the distribution side, which can provide ancillary services to system operators through power flexibility aggregation. Therefore, accurately evaluating the aggregate flexibility of EVs in a charging station is essential. However, EV owners participating in flexibility aggregation programs often introduce scheduling deviation uncertainty, where actual arrival and departure times may deviate from the scheduled ones, complicating the evaluation process and compromising its accuracy. To address this issue, this paper proposes a learning-assisted online aggregate EV flexibility characterization method that integrates an inner-approximation optimization model with a machine learning model. Specifically, an optimization model is developed to represent the aggregate flexibility as an inner-box approximated region. Additionally, a neural network is designed to learn and estimate the bound correction coefficients from historical data, which are then used to adjust the upper and lower bounds of flexibility affected by scheduling deviation uncertainty. Case studies demonstrated the improved accuracy performance achieved after applying the correction.