结合知识图谱与大语言模型的风电装备智能问答系统  

An Intelligent Q&A System for Wind Turbine Equipment Combining Knowledge Graph and Large Language Model(LLM)

作　　者：石致远 张佳蕾 孔志伟 伏洪兵 徐海 王淑营[3] 闫富乾 王立闻 凌乐 SHI Zhiyuan;ZHANG Jiaei;KONG Zhiwei;FU Hongbing;XU Hai;WANG Shuying;YAN Fuqian;WANG Liwen;LING Le(DECAcademy of Science and Technology Co.,Ltd.,611731,Chengdu,China;Dongfang Electric Wind Power Co.,Ltd.,618000,Deyang,Sichuan,China;Southwest Jiaotong University,610031,Chengdu,China)

机构地区：[1]东方电气集团科学技术研究院有限公司,成都611731 [2]东方电气集团东方电气风电股份有限公司,四川德阳618000 [3]西南交通大学,成都610031

出　　处：《东方电气评论》2024年第3期77-84,共8页Dongfang Electric Review

摘　　要：本研究开发了一套专为风电业务人员设计的智能问答系统,该系统融合了结构化知识图谱和经过微调的先进大模型技术,提供高效且准确的信息检索服务,以促进决策效率的提升和企业效益的增长。在深度分析企业风电装备数据的基础上,本文首先采用本体建模、数据处理和知识融合等手段构建了一个风电装备领域知识图谱。随后,本研究采用JointKG-Bert模型对问题进行精确分类和实体识别,并将识别结果填充事先设定的知识库Cypher查询语句模板和大语言模型的Prompt模板,进而实现了知识库的问答与大模型的问答,构筑了一个高效的双轨问答机制。经专家评测表明,问答系统在处理领域问题时知识库问答准确率达到88.7%,经调整的大模型在处理领域问题和非领域问题的准确率分别达到66.2%和91.25%,满足了企业在风电装备信息检索方面的需求。本研究提出的问答系统显著提升了风电行业从业人员的信息获取效率,并为企业提供了一种创新的知识管理和应用解决方案。This study aims to develop an intelligent Q&A system designed for wind power business personnel,which incorporates structured knowledge graphs and fine-tuned advanced big model techniques,with the aim of providing efficient and accurate information retrieval services to facilitate decision-making efficiency and corporate benefit growth.Based on the in-depth analysis of enterprise wind power equipment data,this paper firstly constructs a knowledge graph in the field of wind power equipment by means of ontology modelling,data processing and knowledge fusion.Subsequently,this study adopts the JointKG-Bert model to accurately classify questions and identify entities,and fills the identification results with the pre-set Cypher query statement template of the knowledge base and the Prompt template of the big language model,which in turn realises the Q&A of the knowledge base and the Q&A of the big model,and constructs a highly efficient dual-track Q&A mechanism.The expert evaluation shows that the Q&A system achieves 88.7%accuracy of the knowledge base Q&A when dealing with domain questions,and the adjusted large model achieves 66.2%and 91.25%accuracy when dealing with domain and non-domain questions,respectively,which meets the enterprise's needs for information retrieval of wind power equipment.The Q&A system proposed in this study significantly improves the information acquisition efficiency of wind power industry practitioners and provides an innovative knowledge management and application solution for enterprises.

关 键 词：风电装备 知识图谱智能问答 JointKG-Bert 大语言模型 

分 类 号：F426.61[经济管理—产业经济]