基于知识图谱与图注意力网络的SG-CIM模型映射技术  

An SG-CIM model mapping technology study via knowledge graph and graphattention network

作　　者：李杏 任小伟 楼轶维 高士杰 葛鑫亮 廖小琦 LI Xing;REN Xiaowei;LOU Yiwei;GAO Shijie;GE Xinliang;LIAO Xiaoqi(Big Data Center of State Grid Corporation of China,Beijing 100053,China;School of Computer Science,Peking University,Beijing 100871,China;Beijing Zhongdian Puhua Information Technology Co.,Ltd.,Beijing 100085,China)

机构地区：[1]国家电网有限公司大数据中心 [2]北京大学计算机学院 [3]北京中电普华信息技术有限公司

出　　处：《科技导报》2023年第15期124-132,共9页Science & Technology Review

摘　　要：为了更加智能化地实现国家电网公司设计的公共数据模型(SG-CIM)的业务需求和提高图谱数据本身的质量和共享程度,提出一种基于图注意力网络的图结构映射模型。首先基于SG-CIM模型的数据构建SG-CIM模型知识图谱和数据库表知识图谱,然后通过图注意力网络分别学习每个图中的实体嵌入,将实体嵌入统一向量空间,最后基于实体向量之间的距离计算相似度,得到2个图的图结构映射结果。实验表明,本文提出的模型在SG-CIM数据模型自动映射方面得到相当不错的结果。In order to intelligently meet the business requirements of the SG-CIM model designed by the State Grid Corporation of China and improve the quality and sharing of the graph data,this paper proposes a graph structure mapping model based on gaph attention network(GAT).First of all,using the data of SG-CIM model,an SG-CIM knowledge graph and a database table knowledge graph are constructed.The entities embedding in each graph are learned separately through the GAN,which are then embeded into a unified vector space.Finally,the graph structure mapping results of the two graphs are obtained after calculating the similarities based on the distances between the entity vectors.Experiments show that the proposed model achieves good results in automatic mapping of SG-CIM model.

关 键 词：SG-CIM模型 知识图谱 图注意力网络 知识图谱映射 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术] TM73[自动化与计算机技术—计算机科学与技术]