Due to the large number of access nodes, complex equipment levels and strong asynchrony of multi-source data in distributed energy grid, traditional monitoring methods are difficult to meet the requirements of high-precision perception and low-delay response at the same time. This paper proposes a device-level real-time data fusion method for augmented monitoring. This paper focused on IEEE 1588 time synchronization, sliding window resampling, quality-aware unified representation, Conv1D local feature extraction, GRU state modeling, Softmax adaptive weighted fusion, edge-side lightweight gated temporal network inference, ring buffer queue and cloud-edge parameter synchronization mechanism. A closed-loop computation link from data alignment, state estimation to local execution output is constructed. The experimental results show that in the simulation scene composed of 8 photovoltaic inverters, 5 smart meters and 12 groups of sensors, the monitoring accuracy of the proposed method reaches 96.5%, the F1-score reaches 95.7%, the average response delay is reduced to 54 ms, and the F1-score of 84.6% is still maintained under the condition of 20% noise and 20% packet loss. It shows that the proposed method can provide reliable support for high-precision real-time perception, edge collaborative processing and intelligent decision-making of distributed energy grid.
