State Perception and Collaborative Evaluation of Active Distribution Networks Based on Multi-Source Data Fusion in Dynamic Edge Scenarios
Abstract
To address multi-source heterogeneity, time-varying topology, constrained communications, and rapid generation-load fluctuations in active distribution networks with widespread integration of distributed photovoltaics, flexible loads, and edge sensing terminals, this study proposes an Edge-Model-Master Collaborative State Perception Framework (EMM-CSPF) for dynamic edge scenarios. The framework jointly models anomaly perception, generation-load uncertainty perception, topology dynamics perception, and state estimation. The edge side performs measurement cleaning, missing-value completion, anomaly identification, and local event encoding; the model layer uses a GNN to extract spatial correlations under time-varying topology, an Informer to characterize long-term generation-load dependencies, and a DNN to jointly map multiple perception features; and the master-station layer performs spatial, temporal, and model collaborative evaluation. Simulation experiments on the IEEE 33-bus system and on a 10 kV case constructed from the topology and electrical parameters of an actual distribution network, with operating measurements generated by simulation, yield voltage-magnitude RMSE values of 0.0049 and 0.0052 p.u. and phase-angle RMSE values of 0.00042 and 0.00046 rad, respectively, reducing voltage-magnitude RMSE by 61.11% and 63.38% relative to WLS. Under 20% missing measurements, 10% abnormal measurements, 20% packet loss, and a four-period communication delay, RMSE remains 0.0071, 0.0067, 0.0068, and 0.0063 p.u., respectively. During topology switching, RMSE is 0.0073 p.u., recovery requires two periods, topology consistency reaches 98.6%, end-to-end response time is 0.053 s, and the maximum error coefficient of variation over ten independent runs is 1.041%. The results confirm the accuracy, robustness, physical consistency, and online applicability of EMM-CSPF in complex dynamic edge scenarios.