Article: Hybrid LLM fine-tuning and RAG algorithm for power communication: a weight optimisation model based on dynamic knowledge retrieval Journal: International Journal of Information and Communication Technology (IJICT) 2026 Vol.27 No.87 pp.21 - 40 Abstract: The paper proposes a load optimisation adaptive algorithm DKR-LOAD) for dynamic knowledge retrieval enhancement. Through innovative fusion of retrieval enhancement generation (RAG) and LLM fine-tuning technology, it introduces a dynamic knowledge retrieval module and a weight adjustment model based on self-criticism preference optimisation (SCPO) to achieve real-time knowledge update and parameter optimisation. The experimental results show that on the lEEE PES dataset, the algorithm has an accuracy rate of 92.0%, an Fl score of 0.9l, and a response time of only 600 ms, which is better than baselines such as GPT-3 (82.5% accuracy). In the robustness test, the noise variance of the algorithm is as low as 0.01, and the practicality expert score is 4.5 points, which proves that it can effectively semantic understanding accuracy, anti-interference ability and practical application value. It offers a scalable solution for power communication. Inderscience Publishers - linking academia, business and industry through research

Title: Hybrid LLM fine-tuning and RAG algorithm for power communication: a weight optimisation model based on dynamic knowledge retrieval

Authors: Qun Xie; Xiangcheng He; Rui Zhong; Shaolei Chen; Yang Yang; Long Tang

Addresses: Power Dispatch Control Center, State Grid Yibin Electric Power Supply Company, Yibin, Sichuan, China ' Power Dispatch Control Center, State Grid Yibin Electric Power Supply Company, Yibin, Sichuan, China ' Power Dispatch Control Center, State Grid Yibin Electric Power Supply Company, Yibin, Sichuan, China ' Power Dispatch Control Center, State Grid Sichuan Electric Power Company, Chengdu, Sichuan, China ' Power Dispatch Control Center, State Grid Yibin Electric Power Supply Company, Yibin, Sichuan, China ' Power Dispatch Control Center, State Grid Yibin Electric Power Supply Company, Yibin, Sichuan, China

Abstract: The paper proposes a load optimisation adaptive algorithm DKR-LOAD) for dynamic knowledge retrieval enhancement. Through innovative fusion of retrieval enhancement generation (RAG) and LLM fine-tuning technology, it introduces a dynamic knowledge retrieval module and a weight adjustment model based on self-criticism preference optimisation (SCPO) to achieve real-time knowledge update and parameter optimisation. The experimental results show that on the lEEE PES dataset, the algorithm has an accuracy rate of 92.0%, an Fl score of 0.9l, and a response time of only 600 ms, which is better than baselines such as GPT-3 (82.5% accuracy). In the robustness test, the noise variance of the algorithm is as low as 0.01, and the practicality expert score is 4.5 points, which proves that it can effectively semantic understanding accuracy, anti-interference ability and practical application value. It offers a scalable solution for power communication.

Keywords: power communication; large language models; LLM fine-tuning; retrieval enhancement generation; RAG; dynamic knowledge retrieval; weight optimisation.

DOI: 10.1504/IJICT.2026.155639

International Journal of Information and Communication Technology, 2026 Vol.27 No.87, pp.21 - 40

Received: 09 Jan 2026
Accepted: 18 May 2026
Published online: 07 Aug 2026 *

Title: Hybrid LLM fine-tuning and RAG algorithm for power communication: a weight optimisation model based on dynamic knowledge retrieval

Keep up-to-date