Robustness to multiple simultaneous gross measurement errors

Evaluate the performance of the Hard-Constrained, Probabilistic, Physics-Informed Factor Graph Neural Network (HCP-PINN) under scenarios involving multiple simultaneous gross measurement errors in distribution-system state estimation.

Background

The paper evaluates HCP-PINN under a controlled bad-data scenario in which the active- and reactive-power injection measurements at one bus of the 70-bus Oberrhein network are scaled to three times their true values. HCP-PINN remains accurate in this single-corruption experiment, whereas weighted least squares is substantially distorted because its measurement weights do not account for gross errors.

The authors explicitly identify the robustness of HCP-PINN under multiple simultaneous gross measurement errors as unresolved. Establishing this behavior would determine whether the reported robustness extends from an isolated corrupted measurement pair to more demanding bad-data conditions involving several concurrent gross errors.

References

While HCP-PINN shows strong empirical robustness in the considered bad data scenario, its performance under multiple simultaneous gross measurement errors remains to be evaluated.

— Hard-Constrained Probabilistic Factor Graph Neural Network for Distribution System State Estimation under Non-Gaussian Uncertainty  (2609.35246 - Azam et al., 28 Sep 2026) in Section 5, subsection “Robustness to High Noise and Outliers”