摘要
针对当前桥梁基础设计自动化和智能化水平不足的问题,提出一种基于多尾神经网络的新型架构,并开发了相应的桩基智能选型算法。该算法有效应对了训练样本数量少、类型多且分布不均的挑战。通过定义案例样本数据格式,构建了一个包含90%常见桥型的桩基选型样本库,涵盖简支梁、连续梁、矮塔斜拉桥等多种桥型。分析样本库数据特征后,明确桩基选型是一个典型的“组合分类”问题,并设计了一个由嵌入层、特征提取层和决策层组成的多尾神经网络模型,采用分阶段训练技术进行优化。在此基础上,研发了桩基智能选型算法,并将其集成到云服务平台,成功接入桥梁智能设计系统,实现了桥梁桩基的智能化设计。通过在长赣高铁S101省道特大桥项目中的应用测试,结果显示该算法显著提高了桩基设计效率,达到传统方法的4倍,进而使全桥设计效率提升超过70%。本研究为人工智能技术在勘察设计行业的应用提供了一种新的解决方案,特别是在解决“冷启动”问题上具有重要意义。
To address the current lack of automation and intelligence in bridge foundation design,this study proposes a novel architecture based on multi-tail neural networks and developed an intelligent selection algorithm of pile foundation.This algorithm effectively tackles the challenges posed by small sample size,diverse types,and uneven distribution in the training data.By defining the data format for case samples,a pile foundation selection sample library was constructed,covering 90%of common bridge types,including simply supported beams,continuous beams,and low-tower cable-stayed bridges.After analyzing the data characteristics of the sample library,pile foundation selection was identified as a typical“combination classification”problem.A multi-tail neural network model,consisting of an embedding layer,feature extraction layer,and decision layer,was designed and optimized using a staged training technique.Based on this,an intelligent selection algorithm of pile foundation was developed and integrated into a cloud service platform,successfully linked to the intelligent bridge design system,thereby achieving the intelligent design of bridge pile foundations.Through application testing in the Changsha-Ganzhou high-speed railway S101 provincial highway bridge project,the results showed that the algorithm significantly improved the efficiency of pile foundation design,reaching four times the efficiency of traditional methods and thereby improving the overall bridge design efficiency by over 70%.This study provides a new solution for the application of artificial intelligence technology in the surveying and design industry,particularly in addressing the“cold start”problem.
作者
陈瓴
柏华军
刘诗文
王新国
CHEN Ling;BAI Huajun;LIU Shiwen;WANG Xinguo(China Railway Siyuan Survey and Design Group Co.,Ltd.,Wuhan 430063,China)
出处
《铁道标准设计》
北大核心
2025年第3期113-121,共9页
Railway Standard Design
基金
国家重点研发计划项目(2021YFB2600400)
中国铁建股份有限公司科技研发计划项目(2022-A02)。
关键词
桥梁基础
桩基选型
多尾神经网络
组合分类
智能设计
bridge foundation
pile foundation selection
multi-tail neural network
combination classification
intelligent design
