Quality-gated automatic repair and verifiable closed-loop control for anomalous power-grid operational data

Abstract

Reliable automated control depends on operational data that remain usable during communication loss, sensor drift, and stuck values. This paper develops topology-and-quality-aware repair (TQ-Repair), a framework that couples anomalous- data restoration with auditable closed-loop control for supervisory control and data acquisition (SCADA), phasor measurement unit (PMU), and distributed optical-fiber temperature streams. Missing and anomalous samples are merged into a unified mask; temporal, seasonal, and topology-aware estimates are fused, projected onto operating constraints, and assigned sample confidence and segment grades. A quality-gated control layer maps repaired voltage, frequency, loading, and temperature to reactive-power support and active-power curtailment, while actuator bounds, ramp limits, and a one-step model check verify each command before release. On a physics-guided 12-node, 48-channel benchmark with 6,500 five-minute samples, repair normalized root-mean-square error (NRMSE) remains 0.1065-0.1369 at 10%-30% missingness and is 60.0% lower than linear repair under joint corruption. In 2,744 control-replay cases, the repaired-data controller reaches 97.7% command-mode agreement with the clean-data oracle, limits unnecessary actions to 1.86%, and matches the oracle residual violation rate of 0.80%. With a confidence threshold of 0.65, 91.7% of risk cases are released automatically and their command agreement rises to 98.4%; the remainder are held or escalated. The results show that repair is operationally meaningful when integrated with quality gating, constrained action generation, and closed-loop effectiveness verification.