A physics-constrained zonal pseudo-measurement imputation method for active distribution networks under advanced metering infrastructure data missingness

This is the authors' abstract. We don't add key points for this paper.

Fine-grained monitoring of active distribution networks relies on complete measurements at the distribution-transformer level; however, owing to communication-link latency the advanced metering infrastructure (AMI) suffers delayed data arrival, so that random missingness is pervasive across sampling cross-sections, while purely data-driven imputation struggles to guarantee physical consistency, so its outputs cannot directly serve feasibility-sensitive downstream applications. To address this, we propose a Physics-Informed Gated Attention Transformer, abbreviated PIGAT, which performs zonal pseudo-measurement imputation following a "correct-then-propagate" principle: the network is divided into radial sub-zones with tie switches as boundaries, the boundary tie branches are observed by SCADA, and the outgoing tie power flows act as free variables that characterize inter-zone coupling. The framework consists of three mutually coupled mechanisms. First, a confidence-aware gate corrects missing entries before spatial aggregation to suppress missing-value contamination, and a physics-informed impedance attention bias encodes electrical proximity by cumulative branch impedance. Second, a differentiable DistFlow hard-constraint output head with denoising enhancement embeds the branch power-flow equations as an analytical recursion, so that the imputation result strictly satisfies this physical constraint by construction. Third, physical partitioning combined with per-node shared projection decouples the parameter count from the number of nodes N and reduces the per-zone node scale, balancing scalability and parallel-training efficiency. Under scenarios with different missing rates, noise levels, and DER penetrations, zonal pseudo-measurement imputation experiments are conducted on the modified IEEE 33-/123-bus systems, validating the synergistic advantages of the proposed method in accuracy, physical consistency, and scalability.