基于机器学习的桥梁工程零件知识库信息动态匹配算法  

Dynamic matching algorithm of bridge engineering parts knowledge base information based on machine learning

作　　者：周恩先 冯剑敏 袁婷 Zhou Enxian;Feng Jianmin;Yuan Ting

机构地区：[1]上海振华重工(集团)股份有限公司,上海200125 [2]中国交通信息科技集团有限公司,北京101300

出　　处：《起重运输机械》2025年第6期82-87,共6页Hoisting and Conveying Machinery

基　　金：上海市启明星项目资助“机器学习算法及BIM技术在钢结构桥梁制造中的应用研究”(22YF1448700);上海振华重工(集团)股份有限公司科研项目“基于图像语义分析的钢结构图纸解析技术研究”(9908000436)。

摘　　要：在整理桥梁工程零件信息时,由于知识库子旁侧序列信息的影响,知识库与实体关系的匹配稳定性较低,导致知识库信息动态匹配过程的适应能力较差。为适应复杂度较高的多种类型信息匹配需求,文中提出了基于机器学习的桥梁工程零件知识库信息动态匹配算法。通过运用迭代自组织数据分析动态聚类方法,聚类桥梁工程不同结构、不同类别零件的多结构的动态信息,构建桥梁工程零件信息知识库;然后,在零件知识库中利用软件定义网络划分知识库中各类别的知识库子旁侧序列信息,形成能够描述零件特性参数或属性的子序列,并生成序列集;将序列集输入改进的长短期记忆网络模型中,通过引入注意力机制提取不同序列向量,引入曼哈顿距离,计算零件信息向量之间的距离,依据该距离确定序列向量之间的相似度,依据相似度结果扩展适应能力,完成桥梁工程零件知识库信息动态匹配。实验结果表明:该算法的动态聚类效果较好,规范化簇类方差值均低于0.03;能够计算不同信息类别、不同序列长度的信息向量的相似度;可依据相似度计算结果确定零件的最佳尺寸信息匹配结果,适应能力较强,满足桥梁工程零件装配需求。When organizing information related to bridge engineering parts,the matching stability between the knowledge base and entity relationships is often compromised due to the influence of sequence information on the knowledge base’s sub-sides.This instability results in poor adaptability of the dynamic matching process of knowledge base information.To address this challenge and meet the requirements for matching various types of highly complex information,a dynamic information matching algorithm based on machine learning is proposed specifically for the knowledge base of bridge engineering parts.By adopting the iterative self-organizing data analysis dynamic clustering method,the dynamic information of multiple structures related to different structures and categories of bridge engineering parts was clustered to construct a knowledge base for bridge engineering parts information.Subsequently,within the parts knowledge base,the sub-side sequence information of each category in the knowledge base was divided through a software-defined network,generating sub-sequences that can be used to describe the characteristic parameters or attributes of the parts and creating sequence sets which were then fed into an improved long short-term memory network model.By introducing an attention mechanism,different sequence vectors were extracted.The Manhattan distance was incorporated to calculate the distances between the parts information vectors.Based on these distances,the similarity between sequence vectors was determined.The adaptability was enhanced by expanding similarity results,thereby completing the dynamic matching of information in knowledge base of bridge engineering parts.

关 键 词：机器学习 桥梁工程零件 知识库 动态匹配算法 动态聚类 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]