Bayesian Active Learning-Based Soft Data Space Calibration for System-Wise Aggregate Flexibility Characterization

There have been growing interests in characterizing system-wide aggregate flexibility to support transmission-side ancillary services and promote large-scale integration of distributed energy resources (DERs). However, accurately characterizing the power flexibility region (PFR) at the substation interface is computationally challenging due to the heterogeneity of the DER devices and the need to satisfy network constraints, resulting in a nonlinear and unpredictable geometry. Therefore, this paper proposes a network-aware, near-optimal, and sample-efficient data-driven method. First, a bus-level PFR estimation model is proposed to aggregate local DER devices at each bus into an inner-box approximated region. Second, a feeder-level PFR estimation model is proposed to aggregate all bus-level PFRs into an outer-box approximated region. A Bayesian active learning-based soft calibration strategy is proposed to iteratively and interactively learn a decision boundary of the binary classifier as an improved feeder-level PFR while providing trustworthy measurement to system operators. Case studies on a small, illustrative IEEE system and modified IEEE 123 bus test feeder validate effectiveness of the proposed methods.