Data-Driven Voltage Constraints for Aggregator Dispatch Under Partial Visibility: A Quantile Regression Approach

Résumé fourni par la source

Distribution system operators (DSOs) impose conservative, topology-blind active-power export bounds to guarantee voltage feasibility, but such bounds lead to unnecessary photovoltaic (PV) curtailment. This paper proposes a datadriven approach in which the aggregator exploits its partial voltage visibility to identify linear voltage surrogates via quantile regression (QR) and enforces them as dispatch constraints. The quantile formulation controls the underapproximation budget on training data; the out-of-sample calibration gap is characterized. QR is compared against ordinary least squares (OLS) and Ridge regression baselines with out-of-sample validation using pinball loss and calibration rate. Simulations on the IEEE 33-bus and 69bus radial feeders under three visibility levels (100%, 80%, 60%) and three underapproximation budgets (ϵ∈ {0.08,0.06,0.05}) show that QR achieves a significantly higher injection cap on the 33-bus feeder (+20–23percentage points), while on the 69bus feeder the gains are configuration-dependent. Notably, OLS at full visibility can yieldlowerutilization than the baseline, demonstrating that the conservative quantile bias is essential. The analysis also reveals that under partial visibility, the AC powerflow bisection, rather than the surrogate, becomes the binding constraint. © IEEE. Accepted manuscript. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media, including reprinting/republishing this material for advertising or promotional purposes, creating new collective works, for resale or redistribution to servers or lists, or reuse of any copyrighted component of this work in other works.