基于人工智能的航空发动机知识智能问答系统  

Intelligent Q & A System for Aircraft Engine Based on Large Language Model

作　　者：车艳鑫 常鸿雯 李阳 CHE YAN-xin;CHANG Hong-wen;LI Yang(AECC Shenyang Engine Research Institute,AECC,Shenyang 100000,China)

机构地区：[1]中国航发沈阳发动机研究所,辽宁沈阳110000

出　　处：《航空计算技术》2025年第2期115-120,共6页Aeronautical Computing Technique

基　　金：航空发动机专项项目资助。

摘　　要：随着航空技术的不断发展,对航空发动机的快速研发迭代提出新的要求,航空发动机相关知识的高效、快速获取和使用的形式有待进一步创新。以航空发动机知识的获取和使用为需求导向,以大规模预训练语言模型技术为基础,构建航空发动机知识智能问答系统。基于以上需求,以航空发动机情报知识为基础,结合机器自动标注和人工标注的方法构建小样本高质量训练语料;选择ChatGLM2-6B作为基座模型;按照“优质语料+预训练语言模型+微调”的研究思路,构建航空发动机知识问答大模型。数据优化、采用检索增强生成技术等策略以10%~30%的幅度有效降低大模型幻觉率,并有效提高大模型的语义相似度。With the continuous development of aviation technology,new requirements have been put forward for the rapid research and development iteration of aviation engines,which requires further innovation in the supply form of knowledge related to aviation engines.Taking knowledge services as the demand orientation,based on cutting-edge big language model technology,this paper aims to promote knowledge reading and Q&A services for aircraft engine research and development,and construct an intelligent Q&A system for aircraft engine knowledge.Based on the requirement analysis of development tasks,the knowledge of aviation engines is divided into different themes,forming two downstream tasks of knowledge object recognition and knowledge question answering models.By combining machine automatic annotation and manual annotation methods,high-quality training corpus with small samples is constructed;ChatGLM2-6B is selected as the basic model.Following the research approach of"high-quality corpus+pre trained large models+fine-tuning",a large model for aircraft engine knowledge Q&A is constructed.Data optimization and the use of retrieval enhancement generation technology effectively reduce the illusion rate of large models by 10%to 40%and improve the semantic similarity of large models.

关 键 词：大语言模型 人工智能 生成式AI 航空发动机 智能问答 

分 类 号：TP31[自动化与计算机技术—计算机软件与理论]