面向测井领域的多模态知识图谱构建  

Construction of Multi-modal Knowledge Graph for Logging Field

作　　者：曹茂俊[1] 林世友 肖阳 王瑞芳 邱斌鑫 CAO Mao-jun;LIN Shi-you;XIAO Yang;WANG Rui-fang;QIU Bin-xin(School of Computer and Information Technology,Northeast Petroleum University,Daqing 163318,China)

机构地区：[1]东北石油大学计算机与信息技术学院,黑龙江大庆163318

出　　处：《计算机技术与发展》2024年第9期195-201,共7页Computer Technology and Development

基　　金：黑龙江省自然科学基金(LH2019F004);中石油科技技术开发项目(2021DJ4001)。

摘　　要：针对测井解释过程中数据多源异构、数据间难以互补融合,不能很好应用于风险评估、解释评价和决策知识提供等问题,提出了一种面向测井领域的多模态知识图谱构建方法。该方法从测井角度出发,采用自顶向下的方式将知识整理分类为通用知识、区域知识和辅助知识等,结合测井解释过程中文本、图片、音视频等多模态资料深入挖掘实体属性关系,搭建了测井领域本体层,并基于CasRel实体关系联合抽取,余弦相似度多模态知识融合和TransR多模态表示学习技术完成了测井领域多模态知识图谱的构建。通过大庆测试服务分公司现场实际验证表明,基于该文构造的测井领域多模态知识图谱有效增强了测井知识的整合、互联和共享。In order to solve the problems of heterogeneous data sources,difficult complementarity and fusion between data,which cannot be well applied to risk assessment,interpretation evaluation and decision-making knowledge provision,a multi-modal knowledge graph construction method for logging field is proposed.From the perspective of logging,the proposed method classifies knowledge into general knowledge,regional knowledge and auxiliary knowledge in a top-down way.By combining multi-modal data such as Chinese text,pictures,audio and video in logging interpretation process,the entity attribute relationship is deeply mined,and the ontology layer of logging domain is built,and based on CasRel entity relation joint extraction,cosine similarity multi-modal knowledge fusion and TransR multi-modal representation learning technology,the multi-modal knowledge graph in logging field is constructed.The practical verification of Daqing test service branch shows that the proposed multi-modal knowledge graph can effectively enhance the integration,interconnection and sharing of logging knowledge.

关 键 词：测井 知识图谱 多模态 知识融合 知识表示 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]