Multi-dimensional fault identification method for power grids based on PMU and SCADA data fusion

Recent studies on power grid fault identification have shifted from relying on single data sources toward exploring the synergistic use of multi-source heterogeneous data. This shift originates from the inherently hybrid measurement infrastructure of power grids: supervisory control and data acquisition (SCADA) systems provide extensive steady-state operational parameters and circuit breaker status information, while phasor measurement units (PMUs), with their high sampling rates and precise time synchronization, deliver unprecedented temporal resolution for observing dynamic grid behavior. However, a fundamental contradiction exists between the low refresh rate of SCADA data and the limited deployment density and high cost of PMU installations. To address the challenges of heterogeneous data fusion, we propose a spatiotemporal registration algorithm is proposed and a gated attention fusion (GAF) mechanism is constructed to capture composite fault characteristics that reflect both dynamic system responses and steady-state topological changes. Moreover, a spatiotemporal feature enhancement (STFE) module is introduced to dynamically adjust feature importance according to different fault scenarios. The proposed method is validated through simulations based on actual grid architectures. Results demonstrate that the approach significantly improves average fault identification accuracy - by approximately 12.5% and 7.3% compared to models using only SCADA or PMU data, respectively - and exhibits enhanced robustness under noisy conditions. This work provides a coordinated processing framework for multi-source heterogeneous grid data, demonstrating its feasibility for integration into intelligent fault defense systems and indicating practical engineering value.