Scholarlink

<p>Timely monitoring and prediction of structural response can effectively prevent structural damage. However, the monitoring data of large civil structures such as foundation pits are typically diverse, voluminous, and heterogeneous, thereby posing significant challenges to the development of accurate predictive models. This paper proposes a structural response prediction framework for handling a large amount of complex monitoring data. The framework incorporates a two-stage clustering process and a lightweight Long Short-Term Memory (LSTM) prediction network. The former first combines similar physical quantities through hierarchical clustering, followed by an effective grouping of monitoring points through improved <InlineEquation ID="IEq1"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="13349_2025_987_Article_IEq1.gif" Format="GIF" Height="14" Rendition="HTML" Resolution="72" Type="Linedraw" Width="13" /> </InlineMediaObject> <EquationSource Format="TEX">\(k\)</EquationSource> <EquationSource Format="MATHML"><math> <mi>k</mi> </math></EquationSource> </InlineEquation>-means clustering that takes into account the data characteristics inherent in both temporal and spatial dimensions. The latter uses a customized LSTM network to train a prediction model for the monitoring data in each cluster. The prediction performance of the framework is verified on a large foundation pit with 260 monitoring points. Results demonstrate that the proposed framework could achieve superior performance than the model utilizing only <i>k-</i>means clustering and the model devoid of any clustering in terms of prediction accuracy and speed.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Structural response prediction of large-scale foundation pit based on multi-stage clustering and LSTM

  • 期刊论文

  • 发表日期：

    2025年7月4日

查看全文
  • Jiazeng Shan,
  • Fengrui Yang,
  • Chaobo Zhang

摘要

Timely monitoring and prediction of structural response can effectively prevent structural damage. However, the monitoring data of large civil structures such as foundation pits are typically diverse, voluminous, and heterogeneous, thereby posing significant challenges to the development of accurate predictive models. This paper proposes a structural response prediction framework for handling a large amount of complex monitoring data. The framework incorporates a two-stage clustering process and a lightweight Long Short-Term Memory (LSTM) prediction network. The former first combines similar physical quantities through hierarchical clustering, followed by an effective grouping of monitoring points through improved \(k\) k -means clustering that takes into account the data characteristics inherent in both temporal and spatial dimensions. The latter uses a customized LSTM network to train a prediction model for the monitoring data in each cluster. The prediction performance of the framework is verified on a large foundation pit with 260 monitoring points. Results demonstrate that the proposed framework could achieve superior performance than the model utilizing only k-means clustering and the model devoid of any clustering in terms of prediction accuracy and speed.