摘要
铁路行业作为传统基建行业,其业务智能化程度尚显不足,数据利用率及共享程度亟需提高。为应对上述挑战,提高铁路行业数据利用效率,推动铁路办公系统智能化升级,探索了基于检索增强生成(RAG)和大语言模型(LLM)的智能检索方法,建立了权限知识库体系。以出差业务为基准测试案例,基于思维链(COT)工程策略实现了企业业务“对话式办理”,显著简化了业务操作流程。结果表明,提出的铁路智能助手系统(RIAS)在问答任务上表现优越,命中率达到91.55%,平均倒数排名(MRR)达到72.64%。相较于联网模式下的Qwen和DeepSeek R1模型,RIAS在专业问题解答上表现显著更好。经验证,基于COT的业务智能办理技术可以有效驱动出差申请这类标准化流程,为企业内业务入口多、业务繁琐等问题提供智能化解决方案。
The railway industry,as a traditional infrastructure sector,exhibits significant deficiencies in business intelligence,with data utilization and sharing urgently needing improvement.To address these challenges and enhance data efficiency while promoting intelligent transformation of railway office systems,this study explores an intelligent retrieval method based on Retrieval-Augmented Generation(RAG)and Large Language Model(LLM),establishing a permission-based knowledge base system.Taking business travel operations as the benchmark test case,the implementation of Chain-of-Thought(COT)engineering strategy enables"conversational processing"of enterprise services,significantly streamlining business operation processes.Experimental results demonstrate that the proposed Railway Intelligent Assistant System(RIAS)achieves superior performance in Q&A tasks,with a hit rate of 91.55%and Mean Reciprocal Rank(MRR)of 72.64%.Compared with Qwen and DeepSeek R1 models in online mode,RIAS shows significantly better performance in professional problem-solving.Verification confirms that COT-based intelligent business processing technology can effectively drive standardized workflows,such as travel application procedures,providing smart solutions for challenges such as multiple service entry points and operational complexity within enterprises.
作者
陈宇文
马俊
CHEN Yuwen;MA Jun(Digital Intelligence Business Unit,China Railway Siyuan Survey and Design Group Co.,Ltd.,Wuhan,Hubei 430063,China)
出处
《自动化应用》
2025年第22期77-85,89,共10页
Automation Application
基金
中国铁建股份有限公司地下空间利用领域科研计划项目“深部地下空间智能化运维管理平台”(2024-W19)。
