基于GCN⁃LSTM组合模型的基坑钢支撑轴力时空序列预测∗
作者:
作者单位:

上海大学力学与工程科学学院,上海 200444

作者简介:

秦世伟(1973—),男,讲师,硕导,博士。主要从事岩土工程检测方向研究。E-mail:10002358@shu.edu.cn

通讯作者:

中图分类号:

TU17

基金项目:

上海市 2021 年度“科技创新行动计划”社会发展科技攻关项目(21DZ1204202)资助


Spatiotemporal Sequence Prediction of Axial Force in Foundation Pit Steel Supports Based on GCN-LSTM Combined Model
Author:
Affiliation:

School of Mechanics and Engineering Science, Shanghai University, Shanghai 200444 , China

Fund Project:

  • 摘要
  • |
  • 图/表
  • |
  • 访问统计
  • |
  • 参考文献
  • |
  • 相似文献
  • |
  • 引证文献
  • |
  • 资源附件
  • |
  • 文章评论
    摘要:

    基坑钢支撑的轴力变化是反应基坑中内力变化的重要指标,也是基坑工程灾害防治的重点研究对象。由于土体力学性质的复杂性以及受力演化的不确定性,单纯通过监测和计算难以把握基坑中实际的内力变化趋势。已有研究表明支撑轴力的演化具有典型的时序特征,可使用时间序列预测模型对数据进行预测分析,但预测精度普遍不高。基坑中多个点位的支撑轴力变化往往具有明显的空间相关性,但现有的模型无法捕捉空间信息。为解决上述问题,使用图卷积神经网络(Graph Convolutional Neural Network, GCN)和长短期记忆网络(Long Short-Term Memory, LSTM),组合构建了能捕捉数据时间和空间特征的时空序列预测模型。该模型根据实际点位的空间信息构建了邻接矩阵并生成对应的空间特征,以支撑轴力,空间信息,温度作为输入特征,来预测支撑轴力的发展趋势。使用上海某车站项目中四个具有空间相关性的点位数据进行预测分析,并将组合模型的预测结果与实测数据、单一LSTM模型预测数据进行对比,结果表明:(1)组合模型的收敛速度更快,对于长周期的数据拟合能力更强,并且能更好的反应数据的波动性;(2)组合模型的精度高于仅考虑时间序列特征的单一LSTM模型,有效提高了支撑轴力数据的预测精度。该模型可为实际工程数值预测提供计算参考。

    Abstract:

    The variation of axial force in foundation pit steel supports is a crucial indicator that reflects the changes in internal force within foundation pit. It is a key research focus in disaster prevention and mitigation in foundation pit engineering. Due to the complexity of soil mechanical properties and the uncertainty of stress evolution, it is challenging to capture the variation of internal force in foundation pit solely through monitoring and calculation. Previous studies have shown that the evolution of support axial force exhibits typical temporal characteristics, suggesting that temporal sequence models can be used for predictive analysis. However, the accuracy of these predictions is generally low. Additionally, significant spatial correlations exist in the axial force variation at multiple locations in founda-tion pit, yet existing models often fail to capture spatial information. To address the issues above, this study establishes a model that combines Graph Convolutional Neural Network (GCN) and Long Short Term Memory (LSTM), which is a spatiotemporal prediction model capable of capturing both temporal and spatial features of the data. The model constructed an adjacency matrix based on the spatial information of actual locations and generated corresponding spatial features. By utilizing axial force, spatial information, and temperature as input parameters, this model predicted the trend of support axial force. Predictive analysis was conducted using data from four spatially correlated points in a metro station in Shanghai, and the results of the combined model were compared with actual measurements and predictions based on the LSTM model. The results indicated that: (1) The combined model exhibited a faster convergence speed, stronger fitting capabilities for long-period data, and better responsiveness to data volatility; (2) The accuracy of the combined model surpassed that of the LSTM model that only considered temporal sequence. The combined model effectively enhances the prediction accuracy of axial force data, providing computational references for numerical predictions in engineering applications.

    参考文献
    相似文献
    引证文献
引用本文

秦世伟,朱则匀,戴自立.基于GCN⁃LSTM组合模型的基坑钢支撑轴力时空序列预测∗[J].防灾减灾工程学报,2024,44(6):1257-1264

复制
分享
文章指标
  • 点击次数:
  • 下载次数:
  • HTML阅读次数:
  • 引用次数:
历史
  • 收稿日期:2023-12-08
  • 最后修改日期:
  • 录用日期:
  • 在线发布日期:2025-01-13
  • 出版日期: