Intelligent Distribution-Network Fault Handling via Cross-Modal Semantic Fusion and Knowledge-Graph Reasoning

Power distribution network fault handling imposes stringent requirements on both response timeliness and decision reliability. The high penetration of distributed renewable generation, the widespread deployment of power-electronic devices, and increasing operating-condition fluctuations further intensify the complexity and uncertainty of fault mechanisms. To address the challenges of cross-validating multi-source heterogeneous information, structurally invoking fault-handling knowledge, and balancing latency and accuracy under resource constraints, this paper develops a cross-modal semantic fusion and knowledge-graph reasoning–based fault-handling system, termed CM-SKG. The system takes synchronous phasor measurement unit (PMU) electrical measurements and UAV dual-spectrum inspection images as its primary information sources, and establishes a closed-loop architecture that integrates semantic sensing, transmission, and reasoning. Specifically, task-oriented semantic representations and controllable compression are performed at the edge device; at the edge server, a learnable semantic alignment matrix and a bidirectional interactive attention mechanism are employed for cross-modal fusion, effectively mitigating granularity discrepancies across heterogeneous data and producing consistent fault semantic evidence; at the control center, a fault-handling knowledge graph is introduced for reasoning, mapping fused semantics to executable handling actions, while a tunable reasoning-depth mechanism enables smooth switching between fast response and deeper inference. Furthermore, we propose a semantic communication efficiency (SCE) metric that jointly accounts for cross-modal fusion quality, reasoning reliability, and end-to-end latency, and use it to drive the coordinated optimization of compression ratio, reasoning depth, and computational resources. Online policy learning is realized via a distributional soft actor–critic (DSAC) algorithm. Simulation results demonstrate that CM-SKG significantly improves fault-handling accuracy and decision stability while satisfying real-time constraints.