Abstract Real-time measurement data synchronization and quality monitoring in global power main-distribution networks face challenges of dynamic topology shifts, heterogeneous communication delays, and frequent data anomalies. To address this, we propose a hybrid deep learning framework integrating spatiotemporal features. First, a Temporal Fusion Transformer (TFT) extracts multi-granularity temporal features across hourly, daily, and weekly windows. Next, an Edge-enhanced Graph Neural Network (EdgeGNN) embeds spatial topology using edge attributes (e.g., line impedance, status). Finally, a dynamic gating mechanism fuses spatiotemporal representations and adapts to grid reconfigurations. Experiments demonstrate average synchronization delays of 26.7-31.5 ms, anomaly detection F1-scores up to 0.89, and stable quality assessment response at 30.7 ms, validating the framework’s superiority in accuracy and robustness for grid data governance.
Li et al. (Mon,) studied this question.