Query-adaptive graph retrieval-augmented generation for traceable power-quality knowledge reasoning

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

Artificial intelligence is increasingly used to support energy-system operation and engineering decision-making, but its practical value depends on whether generated answers can be grounded in verifiable domain evidence. This issue is acute in power-quality management, where questions about harmonic distortion, voltage deviation, equipment sensitivity, assessment limits, and mitigation measures often require engineers to connect standards, manuals, textual event reports, and historical work-order descriptions rather than retrieve a single document fragment. This paper proposes PQRAG, a query-adaptive graph retrieval-augmented generation(RAG) framework for traceable power-quality knowledge reasoning. PQRAG builds a hybrid vector-graph knowledge base from unstructured power-quality documents, rewrites user queries with domain terminology, adapts graph retrieval to entity-specific and scenario-level intents, and organizes retrieved evidence as relation chains before answer generation. A benchmark containing 3,219 power-quality questions is constructed to evaluate objective-answer accuracy and evidence traceability against representative graph-based RAG baselines. The results show that query-adaptive graph retrieval improves answer accuracy, while ablation studies indicate that deterministic keyword extraction and evidence-chain synthesis are important for reliable generation. More importantly, the framework makes the reasoning basis of AI-assisted responses inspectable by exposing the links among disturbance phenomena, indicators, standard clauses, equipment constraints, and corrective actions. These findings suggest that graph-grounded retrieval can serve as an auditable knowledge layer for trustworthy AI applications in energy-system engineering.