军事知识服务平台
洪涝灾害"空地一体化"应急预案的知识图谱构建方法
A knowledge graph construction method for"air-ground integrated"emergency plans for flood disasters
摘要 [目的]针对洪涝灾害"空地一体化"应急中多源异构信息分散、传统文档式预案难以支撑快速感知与智能响应的问题,提出基于"情景-任务"范式的知识图谱构建方法,实现灾害情景与应急任务的结构化表达、语义关联与任务推理.[方法]引入通用信息抽取模型(UIE)与预定义本体结构化方法,结合大语言模型ChatGLM2-6B设计语义校准模块,建立灾害情景与任务语义的实体识别与关系抽取体系;其次,通过融合RoBERTa动态词向量、Levenshtein距离与图神经网络(GNN)技术,实现应急任务与情景要素的语义对齐与结构统一.最后,基于郑州2021年"7·20"暴雨洪涝灾害事件,构建覆盖灾害事件、应急任务、灾害数据与模型方法的图谱实例,并验证方法的准确性.[结果]①实体抽取模型(UIE+ChatGLM2-6B)在实体识别和关系抽取两个任务中的F1值(精准率P与召回率R的调和平均值)分别达到87.9%和86.8%.②实体融合模型(RoBERTa+Levenshtein+GNN)的平均F1值提升至87.1%.③在模型应用方面,构建的洪涝灾害"空地一体化"应急预案知识图谱覆盖了4329个实体节点与3353类语义关系,实现了郑州"7·20"暴雨洪涝灾害的应急任务辅助决策查询与灾害链式推理.[结论]构建的基于情景驱动的洪涝灾害"空地一体化"应急任务知识表达体系,可突破传统文档式预案在语义融合与任务联动建模中的瓶颈,提升应急预案的结构化与智能化水平,具备良好的通用性与应用前景.
Abstract:
[Objective]In"air-ground integrated"emergency response for flood disasters,multi-source heterogeneous information is fragmented,and traditional document-based emergency plans are unable to support rapid situational awareness and intelligent response.To address this challenge,this study proposes a knowledge graph construction method based on the"scenario-task"paradigm to enable structured representation,semantic association,and task inference between disaster scenarios and emergency tasks.[Methods]A universal information extraction(UIE)model and a predefined entity structuring method were introduced,combined with the large language model ChatGLM2-6B.A semantic calibration module was designed,through which an entity recognition and relationship extraction framework for disaster scenarios and task semantics was established.Subsequently,by integrating RoBERTa dynamic word embeddings,Levenshtein distance,and graph neural network(GNN)techniques,semantic alignment and structural unification of emergency tasks and scenario elements were achieved.Finally,based on the 2021 Zhengzhou"7·20"torrential rain and flood disaster event,a knowledge graph instance covering disaster events,emergency tasks,disaster data,and modeling methods was constructed to validate the accuracy of the approach.[Results](1)The entity extraction model(UIE+ChatGLM2-6B)achieved F1-scores(the harmonic mean of precision(P)and recall(R))of 87.9%and 86.8%in entity recognition and relationship extraction tasks,respectively.(2)The average F1-score of the entity fusion model(RoBERTa+Levenshtein+GNN)increased to 87.1%.(3)In terms of model application,the constructed"air-ground integrated"flood emergency plan knowledge graph comprised 4329 entity nodes and 3353 semantic relations,which enabled auxiliary decision-making queries for emergency tasks and disaster chain reasoning for the Zhengzhou"7·20"torrential rain and flood disaster.[Conclusion]The constructed scenario-driven knowledge representation system for"air-ground integrated"emergency tasks in flood disasters can overcome the limitations of traditional document-based plans in semantic fusion and task linkage modeling.It enhances the structural and intelligent level of emergency planning and demonstrates strong generalizability and application potential.
刊 名 资源科学 ISTICPKU
年,卷(期): 2026, 48(9)
机标分类号: TP391 TP181 TN912.34
在线出版日期: 2026年10月21日
基金项目: 72304253:国家自然科学基金22YJC630027:教育部人文社会科学研究项目26HASTIT058:河南省高校科技创新人才支持计划项目,