铁道标准设计2025，Vol.69Issue(8)：72-79,8.DOI:10.13238/j.issn.1004-2954.202306170001
基于图神经网络铁路桥梁主梁推荐算法研究
Research on Recommendation Algorithm for Main Girders of Railway Bridges Based on Graph Neural Networks
摘要
Abstract
With the development of intelligent technologies,the hole layout design of railway bridge spans has been evolving toward integration,digitalization,visualization,and intelligence.Research on the intelligent design of railway bridges,both domestically and internationally,has mainly focused on bridge modeling and collaborative design,while intelligent hole layout design for bridges remains a technical gap.In this context,a main girder recommendation algorithm for railway bridges was developed based on the AGOAM model using graph neural networks,enabling the selection of main girders for control points across the bridge span and providing support for intelligent decision-making algorithms for bridge span schemes.By conducting in-depth research into cutting-edge intelligent recommendation technologies,the AGOAM model was proposed,consisting of a preprocessing layer,subgraph construction layer,node matching layer,graph pooling layer,and graph matching layer.It incorporated internal and external attribute interaction between control points and beam types,along with ontology feature enhancement using an integrated attention mechanism.This enabled the optimization of embedded representations for control points and girder types,as well as their efficient matching using the similarity algorithm.The model's performance on the validation set using AUC,LogLoss,Precision,and NDCG indicators demonstrated that the algorithm achieved good accuracy,ranking ability,and recommendation quality.关键词
铁路桥梁/布孔设计/图神经网络/智能设计/注意力机制/推荐算法/相识度算法Key words
railway bridges/hole layout design/graph neural networks/intelligent design/attention mechanism/recommendation algorithm/similarity algorithm分类
交通工程引用本文复制引用
柏华军,郑洪,陈瓴,桂浩..基于图神经网络铁路桥梁主梁推荐算法研究[J].铁道标准设计,2025,69(8):72-79,8.基金项目
国家重点研发计划项目(2021YFB2600400) （2021YFB2600400）
中国铁建股份有限公司科技研发计划项目(2022-A02) （2022-A02）
