基于知识图谱的EPC管理GNN知识融合方法

曹怀志; 刘兴宁; 来亦姝; 张涵需; 钟千有; 刘强

doi:10.13928/j.cnki.wrahe.2025.S2.001

水利水电技术（中英文） ›› 2025, Vol. 56 ›› Issue (S2) : 1 -5. DOI: 10.13928/j.cnki.wrahe.2025.S2.001

知识驱动的长江大保护智慧EPC管控技术专栏

基于知识图谱的EPC管理GNN知识融合方法

曹怀志 ¹ ,
刘兴宁 ² ,
来亦姝 ³^,⁴ ,
张涵需 ³^,⁴ ,
钟千有 ³^,⁴ ,
刘强 ²

作者信息 +

EPC management GNN knowledge fusion method based on knowledge graph

Huaizhi CAO ¹ ,
Xingning LIU ² ,
Yishu LAI ³^,⁴ ,
Hanxu ZHANG ³^,⁴ ,
Qianyou ZHONG ³^,⁴ ,
Qiang LIU ²

Author information +

文章历史 +

PDF (6648K)

摘要

水利EPC企业在工程项目的全生命周期内积累了丰富的管理和经验知识。为解决企业管理知识难以更新增长，提出基于GNN的水利EPC企业管理知识融合方法。首先明确水利工程项目管理及EPC企业管理的知识架构，并依据相关标准规范与过程文件构建知识图谱的数据层，以形成各模块的专题知识图谱；随后，通过实例命名实体识别技术实现各专题本体的映射；然后，采用图神经网络进行实体对齐，以保证各专题知识的融合；最后以长江大保护项目为案例进行实例验证，并结合EPC项目管理的各个阶段，完成专题知识图谱的融合。结果表明，命名实体识别使用基于RoBERTa-wwm-ext的模型，取得了0.895的F1值，表现出良好的效果。在实体对齐方面，应用AttrGNN方法，其Hits@1和MRR值分别达到了97.6和0.97,显示明显提升。研究结果为提升水利EPC企业的管理效率和推动管理水平的提升提供了新的思路和方法，具有重要的理论和实践意义。

Abstract

Water conservancy EPC enterprises have accumulated rich management and experience knowledge in the whole life cycle of the project. In order to solve the problem that enterprise management knowledge has trouble updating and growing, a knowledge fusion method of water conservancy EPC enterprise management based on GNN is proposed. Firstly, the knowledge architecture of water conservancy project management and EPC enterprise management is analyzed, and the data layer of knowledge graph is constructed according to relevant standard specifications and process documents to form the thematic knowledge graph of each module. Then, the knowledge graph of each topic ontology is mapped by instance named entity recognition technology. Then, the graph neural network is used to align the entities to ensure the integration of thematic knowledge. Finally, the Yangtze River protection project is taken as an example to verify the case, and the integration of thematic knowledge map is completed in combination with each stage of EPC project management. The result show that the named entity recognition uses a model based on RoBERTa-wwm-ext, and achieves an F1 value of 0.895, showing good result. In terms of entity alignment, using the AttrGNN method, its Hits@1 and MRR values reached 97.6 and 0.97, respectively, showing a significant improvement. This study provides new ideas and method for improving the management efficiency of water conservancy EPC enterprises and promoting the improvement of management level, which has important theoretical and practical significance.