摘要
针对洪涝灾害应急管理中需实时更新防汛领域知识库的需求,该文通过LangChain框架结合外部防汛领域知识库与大语言模型,提出了基于实时知识库的洪涝应急决策智能问答模型LR-GLM(LangChain RAG Generative Language Model)。该模型集成了RAG(Retrieval-augmented Generation)技术,通过向量匹配机制并结合微调后的ChatGLM2(Chat Generative Language Model 2)模型生成回答。以湖北省随县洪涝灾害应急演练为案例,采用人工评估和自动评估,验证模型的有效性。结果表明:该模型在多轮回答及复杂决策场景下表现优异,能够快速响应并提供针对现场态势的答案,有效提升了应急智能问答的准确性和实时性,有利于应急指挥团队制定更高效切实可行的洪涝灾害应急方案。
In response to the requirement for real-time updating of the knowledge base in flood disaster emergency management,the external flood prevention knowledge base and large language models are combined through the LangChain framework.An intelligent question-answering model LR-GLM(LangChain RAG Generative Language Model)for flood emergency decision-making based on real-time knowledge repository is proposed.The model incorporates RAG(Retrieval-Augmented Generation)technology and employs a vector matching mechanism combined with a fine-tuned ChatGLM2(Chat Generative Language Model 2)for language generation.Using the question-answer pairs collected from a flood emergency drill in Suixian County,Hubei Province,as a case study,the model's effectiveness is validated through both manual and automatic evaluation methods.The results demonstrate the model's excellent performance in multi-turn dialogues and complex decision-making scenarios by providing rapid and context-specific responses,which will significantly enhance the accuracy and real-time capability of intelligent emergency question-answering,thereby supporting emergency command teams in formulating more efficient and practical flood disaster response strategies.
作者
王喆
王泽晗
黄海辰
李瑞钦
WANG Zhe;WANG Zehan;HUANG Haichen;LI Ruiqin(School of Safety Science and Emergency Management,Wuhan University of Technology,Wuhan 430070,China;China Emergency Management Research Center,Wuhan University of Technology,Wuhan 430070,China)
出处
《灾害学》
北大核心
2025年第4期132-137,共6页
Journal of Catastrophology
基金
教育部人文社会科学研究青年基金项目“城镇防汛应急物资公私协同储备与调度研究”(20YJC630154)
中央高校基本科研业务费专项资金项目“大安全态势感知分析干预应用研究”(104972024DZB0003)。
关键词
洪涝灾害
实时知识库
检索增强生成
大语言模型
智能问答系统
flood disaster
real time knowledge base
retrieval-augmented Generation(RAG)
large language model(LLM)
intelligent question answering system
