Knowledge-graph-enhanced and LLM-guided fault diagnosis for VSC-HVDC systems

Accurate fault diagnosis in voltage-source-converter high-voltage direct-current (VSC-HVDC) systems is vital for grid stability, yet existing data-driven methods often lack interpretability and robustness under complex operating conditions. This study introduces a hybrid diagnostic framework that integrates a knowledge-graph-guided reasoning module with large-language-model (LLM)-based contextual understanding. Event logs and SCADA measurements are first converted into structured knowledge representations, from which a fault-causality graph is constructed to capture inter-event dependencies. The LLM then performs multi-stage reasoning—semantic normalization, intent inference, and consistency verification—to extract failure signatures and infer root causes. Experimental evaluation on real VSC-HVDC operation data demonstrates that the proposed framework improves diagnostic F1-score by more than 10% compared with conventional sequence or attention models, while providing interpretable fault chains aligned with domain knowledge. These results suggest a reproducible pathway toward physics-informed, explainable AI for power-electronic system monitoring.