Low-latency cloud-edge fault diagnosis and location for networked distribution automation systems

Abstract

Networked automation systems require fault-diagnosis functions that are fast enough for local control while still maintaining global model consistency. This paper repackages distribution-feeder fault location as a low-latency cloud-edge automation problem. The edge layer performs event triggering, phasor-sequence feature compression, topology-masked graph neural network (GNN) inference, and first-response section ranking. The cloud layer maintains a digital-twin feeder model, verifies low-confidence cases, and distributes calibrated model updates to field gateways. A confidence-aware split strategy is developed so that routine faults are processed on the edge fast path, whereas difficult topology and distributed energy resource (DER) cases are selectively escalated to the cloud. A case study on a modified Institute of Electrical and Electronics Engineers (IEEE) 33-bus active feeder shows 98.4% section-level accuracy, 0.21-km mean distance error, 118-ms median decision latency, and 31.5-kB uplink payload per event under mixed DER, fault-resistance, and topology-switching scenarios. The results indicate that cloud-edge collaboration can support reliable fault diagnosis and location in networked distribution automation without continuous raw-waveform upload. The same cloud-edge automation framework can also incorporate optical-fiber sensing, infrared inspection, or camera- or light detection and ranging (LiDAR)-based field inspection streams when distribution assets are monitored by optical sensing infrastructure.