Article: Towards dynamic knowledge graph-based RAG system optimisation for power system large language models Journal: International Journal of Information and Communication Technology (IJICT) 2026 Vol.27 No.90 pp.27 - 48 Abstract: To address the limitations of traditional large language models in adapting to the intelligent construction of new power systems, as well as the insufficient recall of domain-specific knowledge, semantic associations, and topological relationships in conventional retrieval-augmented generation (RAG) architectures, this paper constructs a knowledge graph-enhanced RAG system for the power domain. It proposes optimisation methods including domain-adaptive semantic embedding, knowledge graph-enhanced multi-strategy retrieval, and dual-constrained compliance generation. Comparative experiments with traditional RAG models demonstrate that the proposed model outperforms conventional RAG in retrieval precision and hallucination suppression, effectively meeting the requirements of various core business scenarios in new power systems and demonstrating strong engineering value. Inderscience Publishers - linking academia, business and industry through research

Title: Towards dynamic knowledge graph-based RAG system optimisation for power system large language models

Authors: Kaijie Liu; Yun Dong; Ou Pu; Jun Li; Xianfu Liu; Zhongqiang Bao; Shibo Zhang; Jian Ye

Addresses: Digital and Intelligent Operation Center, Guangxi Power Grid Co., Ltd., Nanning Guangxi, 530015, China ' Digital and Intelligent Operation Center, Guangxi Power Grid Co., Ltd., Nanning Guangxi, 530015, China ' Power Grid Planning Research Center, Guangxi Power Grid Co., Ltd., Nanning Guangxi, 530015, China ' Power Grid Planning Research Center, Guangxi Power Grid Co., Ltd., Nanning Guangxi, 530015, China ' Power Grid Planning Research Center, Guangxi Power Grid Co., Ltd., Nanning Guangxi, 530015, China ' Power Grid Planning Research Center, Guangxi Power Grid Co., Ltd., Nanning Guangxi, 530015, China ' Power Grid Planning Research Center, Guangxi Power Grid Co., Ltd., Nanning Guangxi, 530015, China ' Power Grid Planning Research Center, Guangxi Power Grid Co., Ltd., Nanning Guangxi, 530015, China

Abstract: To address the limitations of traditional large language models in adapting to the intelligent construction of new power systems, as well as the insufficient recall of domain-specific knowledge, semantic associations, and topological relationships in conventional retrieval-augmented generation (RAG) architectures, this paper constructs a knowledge graph-enhanced RAG system for the power domain. It proposes optimisation methods including domain-adaptive semantic embedding, knowledge graph-enhanced multi-strategy retrieval, and dual-constrained compliance generation. Comparative experiments with traditional RAG models demonstrate that the proposed model outperforms conventional RAG in retrieval precision and hallucination suppression, effectively meeting the requirements of various core business scenarios in new power systems and demonstrating strong engineering value.

Keywords: retrieval-augmented generation; RAG; power knowledge graph; large language model; LLM; power system intelligence.

DOI: 10.1504/IJICT.2026.155731

International Journal of Information and Communication Technology, 2026 Vol.27 No.90, pp.27 - 48

Received: 03 Apr 2026
Accepted: 13 Jun 2026
Published online: 11 Aug 2026 *

Title: Towards dynamic knowledge graph-based RAG system optimisation for power system large language models

Keep up-to-date