Scholarlink

<p>Structural Health Monitoring (SHM) relies on reliable monitoring data to achieve real-time assessments of structural performance. However, environmental impacts and sensor failures may result in missing SHM data. Current advanced methods utilize deep learning models (e.g., Convolutional Neural Network (CNN) and Graph Convolutional Network (GCN)) to reconstruct missing SHM data by modeling the correlations between complete and incomplete signals. However, these methods often fail to fully exploit the spatiotemporal correlations present in the monitoring data, inadequately capturing key spatial and temporal features, which leads to suboptimal reconstruction performance. This paper proposes the AO-CNN-BiLSTM-Attention data reconstruction method. First, the CNN is combined with Bidirectional Long Short-Term Memory (BiLSTM) for extracting spatial features and capturing temporal correlations from SHM data. Second, Squeeze-and-Excitation and Temporal Attention mechanisms are integrated to adaptively capture critical spatial and temporal features. Third, the Aquila Optimization algorithm is utilized to identify the optimal hyperparameter combination, enhancing the capability of the method in data reconstruction. Strain monitoring data collected from the South Taihu Central Business District Tower is used for demonstrating the proposed method, whose performance is compared with CNN-BiLSTM and GCN-BiLSTM methods. The comparison results show that, under both normal and typhoon conditions, the proposed method reduces the root mean square error, mean squared error, and mean absolute error by average values of 36.51%, 55.60%, and 32.96%, respectively. Compared with the other methods, the proposed method achieves better reconstruction performance and stronger robustness.</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

An approach for reconstruction of missing SHM data using AO-CNN-BiLSTM-Attention

  • 期刊论文

  • 发表日期：

    2026年2月11日

查看全文
  • Hao Chen,
  • Hua-Ping Wan,
  • Zhi-Hai Weng,
  • Ye Zhou,
  • Hong-Chang Xu

摘要

Structural Health Monitoring (SHM) relies on reliable monitoring data to achieve real-time assessments of structural performance. However, environmental impacts and sensor failures may result in missing SHM data. Current advanced methods utilize deep learning models (e.g., Convolutional Neural Network (CNN) and Graph Convolutional Network (GCN)) to reconstruct missing SHM data by modeling the correlations between complete and incomplete signals. However, these methods often fail to fully exploit the spatiotemporal correlations present in the monitoring data, inadequately capturing key spatial and temporal features, which leads to suboptimal reconstruction performance. This paper proposes the AO-CNN-BiLSTM-Attention data reconstruction method. First, the CNN is combined with Bidirectional Long Short-Term Memory (BiLSTM) for extracting spatial features and capturing temporal correlations from SHM data. Second, Squeeze-and-Excitation and Temporal Attention mechanisms are integrated to adaptively capture critical spatial and temporal features. Third, the Aquila Optimization algorithm is utilized to identify the optimal hyperparameter combination, enhancing the capability of the method in data reconstruction. Strain monitoring data collected from the South Taihu Central Business District Tower is used for demonstrating the proposed method, whose performance is compared with CNN-BiLSTM and GCN-BiLSTM methods. The comparison results show that, under both normal and typhoon conditions, the proposed method reduces the root mean square error, mean squared error, and mean absolute error by average values of 36.51%, 55.60%, and 32.96%, respectively. Compared with the other methods, the proposed method achieves better reconstruction performance and stronger robustness.