面向水利防汛抢险的知识图谱构建与应用  

Construction and application of knowledge graph for flood defense andrescue of water infrastructure

作　　者：张栋梁[1,2] 周伟[1,2,3] 马刚[1,2,3] 王旭东[2] 刘禹 王小毛 ZHANG Dongliang;ZHOU Wei;MA Gang;WANG Xudong;LIU Yu;WANG Xiaomao(Institute of Water Engineering Sciences,Wuhan University,Wuhan 430072,China;State Key Laboratory of Water Resources Engineering and Management,Wuhan University,Wuhan 430072,China;Key Laboratory of Rock Mechanics in Hydraulic Structural Engineering of Ministry of Education,Wuhan University,Wuhan 430072,China;CISPDR Corporation,Wuhan 430010,China)

机构地区：[1]武汉大学水工程科学研究院,湖北武汉430072 [2]武汉大学水资源工程与调度全国重点实验室,湖北武汉430072 [3]武汉大学水工岩石力学教育部重点实验室,湖北武汉430072 [4]长江设计集团有限公司,湖北武汉430010

出　　处：《水利学报》2025年第3期341-353,共13页Journal of Hydraulic Engineering

基　　金：国家重点研发计划课题(2022YFC3005505);国家自然科学基金项目(52322907,52179141,U23B20149);中央高校基本科研业务费专项项目(2042024kf1031,2042023kfyq03)。

摘　　要：知识平台是数字孪生水利的重要组成部分,而水利知识分散在多源文本中,具有显著的非结构化与碎片化特征,知识提取与有效利用面临挑战。本文以水利防汛抢险文本为对象,针对领域数据质量偏低、知识利用率不足问题,改进知识抽取模型、融合外源数据,提出了联合非结构化与半结构化数据的防汛抢险知识图谱智能构建方法。首先利用大语言模型提取非结构化文本术语,解析术语主题构建本体模型。采用预训练模块强化文本表示特征,引入卷积模块改进实体知识抽取模型,提出实体数据增强方法提升知识抽取准确率。然后抽取外源百科数据扩充知识面,构建了更完备的防汛抢险知识图谱。文本试验显示,所提模型实体知识抽取F 1值为89.91%,明显优于基线模型。最后介绍了知识图谱在防汛抢险领域的应用方式,可形成数字孪生水利建设的知识引擎,为防汛研判与决策提供知识支撑。The knowledge platform is an important component in digital twin of water conservancy.However,water conservancy knowledge is dispersed across multi-source texts,which exhibit obvious unstructured and fragmented characteristics,and knowledge extraction and effective utilization face challenges.To addresses the issues of low data quality and underutilization of knowledge in the field,this study focuses on the texts of flood defense and emergency rescue,and proposes an intelligent method for constructing a flood defense and emergency rescue knowledge graph by improving the knowledge extraction model and combining unstructured data and external semi-structured data.Initially,a large language model is employed to extract term from unstructured texts and construct an ontology model based on term themes,a pretraining module is used to enhance text representation features,and a convolutional module is introduced to improve the entity knowledge extraction model,and an entity data enhancement method is proposed to improve model accuracy.Then external encyclopedia data is extracted to expand the knowledge coverage to build a complete flood defense and rescue knowledge graph.Experimental results demonstrate that the proposed model achieves an F 1 score of 89.91%in entity knowledge extraction,significantly outperforming baseline models.Finally,the application method of knowledge graph in the field of flood defense and rescue is introduced,which can form a knowledge engine for digital twin of water conservancy construction,providing knowledge support for flood control research and decision-making.

关 键 词：防汛抢险 知识图谱 知识抽取 多源数据 数字孪生 

分 类 号：TV698.2[水利工程—水利水电工程]