华北水利水电大学学报（自然科学版）2025，Vol.46Issue(5)：34-42,9.DOI:10.19760/j.ncwu.zk.2025071
基于大语言模型的抽水蓄能电站智能运维辅助系统构建研究
Intelligent Operation and Maintenance for Pumped Storage Power Stations Based on Large Language Model
摘要
Abstract
[Objective]To address the complexities of operation and maintenance(O&M)work at pumped storage power sta-tions,the high demand for specialized knowledge,the inefficiency of traditional knowledge retrieval,and the difficulty of gen-eral large language models(LLMs)in accurately understanding the unique terminology and complex operating conditions spe-cific to pumped storage O&M,this study aims to enhance the professional competence and O&M efficiency of personnel,im-prove knowledge reuse efficiency,and meet the national strategic requirements for developing new quality productivity in water conservancy and electric power systems by constructing an intelligent O&M assistance system tailored for pumped storage power stations.[Methods]The Pumped Storage-Artificial Intelligence Operation and Maintenance(PS-AIOM)system was con-structed based on large language models.Utilizing a locally deployed LLM as the foundation,the system integrated retrieval-augmented generation(RAG)and prompt engineering techniques.Through external knowledge base retrieval,it enhanced the accuracy of LLM applications in pumped storage O&M.To ensure response accuracy for O&M tasks,a dedicated O&M knowl-edge base was constructed.To validate system effectiveness,a test dataset was created.[Results](1)An evaluation method combining keyword matching and semantic similarity was adopted to score PS-AIOM's responses,considering both the specific matching of content and overall textual similarity for a comprehensive assessment.(2)PS-AIOM achieved a satisfaction rate of 83％for O&M queries,with a comprehensive average score of 0.8138,demonstrating its effectiveness in assisting O&M personnel.(3)PS-AIOM significantly outperformed advanced general LLMs(Model Z:4％satisfaction rate,0.3982 average score;Model T:17％satisfaction rate,0.4775 average score).[Conclusion]By integrating a locally deployed LLM base,RAG technology,and a domain-specific knowledge base,the PS-AIOM system effectively overcomes the limitations of general LLMs regarding professional domain knowledge scarcity.It achieves relatively accurate knowledge application within the O&M domain of pumped storage power stations.关键词
抽水蓄能电站/智能运维/大语言模型/检索增强生成Key words
pumped storage power station/intelligent operation and maintenance/large language model/retrieval-augmented generation分类
建筑与水利引用本文复制引用
胡昊,许昭一,崔争艳,张兴奎,张浩宇..基于大语言模型的抽水蓄能电站智能运维辅助系统构建研究[J].华北水利水电大学学报（自然科学版）,2025,46(5):34-42,9.基金项目
国家自然科学基金项目(52079053) （52079053）
河南省中央引导地方科技发展资金项目(Z20241471035) （Z20241471035）
河南省重点研发专项(241111210300) （241111210300）
河南省自然科学基金项目(252300420056) （252300420056）
河南省高等学校重点科研项目(25B520056). （25B520056）
