摘要
在智慧城市发展进程中,交通系统的精细化管理和智能化服务面临海量异构数据处理的挑战。传统交通信息查询系统存在数据源异构性强、自然语言交互能力不足、长尾查询场景覆盖有限等问题。文章基于ChatGLM3大语言模型,创新性地构建了融合NL2SQL(Natural Language to Structured Query Language)技术的智能问数系统,通过动态Schema对齐、LoRA微调优化及多维度提示工程技术,实现了交通领域复杂自然语言查询到精准SQL指令的智能转换。实验结果表明,经过微调的模型在交通信息查询任务中准确率达到78.9%,较基线模型提升15.8个百分点。本研究为交通管理智能化转型提供了创新技术路径,并对大模型在垂直领域的深度适配进行了系统性探索。
In the process of smart city development,the refined management and intelligent services of the transportation system are confronted with the challenge of processing massive heterogeneous data.The traditional traffic information query system has problems such as strong heterogeneity of data sources,insufficient natural language interaction ability,and limited coverage of long-tail query scenarios.Based on the ChatGLM3 large Language model,this paper innovatively constructs an intelligent Query system integrating NL2SQL(Natural Language to Structured Query Language)technology.Through dynamic Schema alignment,LoRA fine-tuning optimization and multi-dimensional prompt engineering technology,the intelligent conversion from complex natural language queries to precise SQL instructions in the transportation field has been achieved.The experimental results show that the fine-tuned model achieves an accuracy rate of 78.9%in the traffic information query task,which is 15.8 percentage points higher than that of the baseline model.This research provides an innovative technical path for the intelligent transformation of traffic management and conducts a systematic exploration of the deep adaptation methods of large models in vertical fields.
作者
沈宇
黄卫东
叶文武
Shen Yu;Huang Weidong;Ye Wenwu(Nanjing University of Posts and Telecommunications,Nanjing 210003,China;Information Industry Development Strategy Institute,Nanjing 210003,China;Chinatelecom Hongxin Information Technology Co.,Ltd.,NanJing 210000,China)
出处
《信息通信技术》
2025年第3期29-36,76,共9页
Information and communications Technologies
关键词
自然语言处理
交通数据分析
大语言模型
NL2SQL
Natural Language Processing
Traffic Data Analysis
Large Language Model
Natural Language to Structured Query Language
作者简介
沈宇,硕士,工程师,主要研究方向为人工智能、大数据、行业智能化应用。;叶文武,硕士,工程师,主要研究方向为计算机视觉、多模态大模型。;黄卫东,博士,教授,南京邮电大学管理学院院长,信息产业发展战略研究院院长,主要研究方向为人工智能、信息产业战略。
