Telemetry-Only Voltage Sensitivity Estimation for Interoperable DER Aggregation

Abstract

Voltage-to-power sensitivity information is a key enabler of voltage regulation, distributed energy resource (DER) dispatch, and inverter-based control in distribution systems with high photovoltaic (PV) penetration. However, practical operators, such as aggregators and virtual power plants, often lack access to full-feeder models or feeder-wide phasor measurement unit (PMU) infrastructure. This study proposes a telemetry-only deep-learning framework to estimate voltage sensitivities using interoperability-enabled DER telemetry without assuming feeder-wide PMU deployment. The method first learns a telemetry-to-voltage surrogate, and then calibrates its Jacobian using sparse finite-difference teacher sensitivities from representative operating points. Thus, the estimator reconstructs the voltages while preserving the local voltage-sensitivity structure required for operational use. The framework was validated on three feeders spanning near-linear to strongly nonlinear voltage behaviors—the IEEE 37-bus, IEEE 123-bus, and IEEE European Low-Voltage (LV) test feeders—using six years of irradiance-driven time-series simulations coordinated through a MATLAB–OpenDSS COM interface. The proposed zero-PMU estimator was compared with PMU-less baselines across these feeders (Case 1); the robustness to PV penetration was examined (Case 2); and the cost–accuracy trade-off was evaluated by adding some strategically placed PMUs (Case 3). The proposed method shows a linearity-dependent advantage that grows with feeder nonlinearity—competitive on the near-linear 37-bus feeder, while reducing the sensitivity root mean square error by approximately 4.3 times over the best linear baseline on the nonlinear LV feeder—and sustains this accuracy across penetration levels, whereas limited PMU augmentation yields further deployment-stage-dependent gains.