基于知识增强的配电网运行信息关系抽取方法  

Relation extraction of distribution network operation information via external knowledge enhancement

作　　者：李强 庄莉 赵峰 王秋琳 张晓东 LI Qiang;ZHUANG Li;ZHAO Feng;WANG Qiulin;ZHANG Xiaodong(State Grid Information&Telecommunication Group Co.,Ltd.,Beijing 102211,China;Fujian Yirong Information Technology Co.,Ltd.,Fuzhou 350001,China)

机构地区：[1]国网信息通信产业集团有限公司,北京102211 [2]福建亿榕信息技术有限公司,福建福州350001

出　　处：《现代电子技术》2024年第5期171-175,共5页Modern Electronics Technique

基　　金：国网信通产业集团两级协同研发项目:新型电力系统国家新一代人工智能开放创新平台研发(SGITYLYRSCJS2310305)科技项目资助。

摘　　要：知识图谱可辅助现场人员处理配电网运行过程中积累的海量异构文本,挖掘高价值运行信息。针对知识图谱构建需求人工标注成本过高的问题,提出一种基于知识增强的远程监督关系抽取方法。利用GloVe编码与Bi-GRU网络对配电网运行语料库的句包进行向量表征;同时,新增外部知识增强模块,通过Attention-GCN模型获取编码文本的句法结构以及非线性关系,实现低成本的关系抽取,解决传统远程监督学习存在的长尾分布及噪声问题,提升关系抽取精度。实验结果表明,该方法在配电网运行信息关系抽取任务中表现优秀,相较于主流模型在精确率上提升6%。Knowledge graph can assist field personnel to process the massive heterogeneous texts accumulated in the operation process of distribution network and mine high⁃value operation information.In view of the fact that the cost of manual labeling for knowledge graph construction requirements is excessively high,a distant supervision relation extraction method based on knowledge enhancement is proposed.GloVe(global vectors for word representation)decoding and Bi⁃GRU(bidirectional⁃gated recurrent unit)are used to implement the vector representation for the sentence packets of the power grid operation corpus.An external knowledge enhancement module is added to obtain the syntactic structure and nonlinear relationship of the encoded text by the Attention⁃GCN(graph convolutional network)model,so as to realize low⁃cost relationship extraction and solve the problems of long⁃tail distribution and noise in traditional distant supervised learning,and then improve the relation extraction accuracy.Experimental results show that the proposed method performs excellently in the extraction task of distribution network operation information relationship,and improves the accuracy by 6%in comparison with the mainstream model.

关 键 词：关系抽取 远程监督 外部知识增强 配电网运行 知识图谱 GloVe编码 

分 类 号：TN99-34[电子电信—信号与信息处理]