深圳大学学报（理工版）2026，Vol.43Issue(5)：591-601,11.DOI:10.3724/SP.J.1249.2026.05591
基于BiLSTM的京沪高铁桥墩沉降时序预测与系统研发
Time-series prediction of pier settlement and system development for the Beijing-Shanghai high-speed railway based on BiLSTM
摘要
Abstract
To improve the accuracy of pier settlement prediction during both the construction and operational stages,we propose a settlement prediction method based on a bidirectional long short-term memory(BiLSTM)network by using the underpass project beneath the Changzhou West Bridge on the Beijing-Shanghai High-Speed Railway as the case study.First,a finite element model is established to simulate the load variations throughout each construction stage and to investigate the settlement behavior of the bridge piers,providing a benchmark for validating the deep learning model.Subsequently,settlement monitoring data collected from piers significantly affected by construction disturbances are preprocessed using exponential smoothing and standardization before being used to develop and optimize the BiLSTM prediction model.After hyperparameter optimization,the proposed model achieves excellent predictive performance on the test dataset,with a mean square error(MSE)as low as 0.006 3.Compared with the finite element approach,the proposed model exhibits significant advantages in computational efficiency,reduced dependency on physical parameters,and fine-grained prediction capability at specific time nodes.In addition,a graphical user interface(GUI)-based prediction system integrating the proposed methods is developed to support intelligent settlement monitoring and efficient operation and maintenance of railway bridges.The proposed approach provides reliable technical support for safety assessment,intelligent settlement early warning of and scientific operation and maintenance of high-speed railway bridge piers under complex construction disturbances.关键词
桥梁工程/高铁桥梁/桥墩沉降/深基坑开挖/有限元/深度学习/双向长短期记忆网络/智能运维Key words
bridge engineering/high-speed railway bridge/pier settlement/deep foundation pit excavation/finite element method/deep learning/BiLSTM/intelligent operation and maintenance分类
建筑与水利引用本文复制引用
胡政汉,王长丹,王景龙,李晓龙,许晶晶..基于BiLSTM的京沪高铁桥墩沉降时序预测与系统研发[J].深圳大学学报（理工版）,2026,43(5):591-601,11.基金项目
National Natural Science Foundation of China(52572500) 国家自然科学基金资助项目(52572500) （52572500）
