Scholarlink

<p>Due to anomalies that occur during data collection, transmission, or storage process in dam health monitoring, deformation data are often incomplete. Accurate imputation of missing deformation data is crucial for assessing dam safety status. While deformation data exhibit spatiotemporal correlations, most existing imputation methods fail to adequately integrate these correlational patterns. To address this issue, this paper proposes a missing deformation data imputation method based on bidirectional spatiotemporal imputation. This method first extracts forward and backward temporal dependency using recurrent neural network, and then integrates spatial relationships among adjacent monitoring points with multilayer perceptron. These imputation results are subsequently combined through an adaptive mechanism. Furthermore, the proposed method is trained within a generative adversarial framework to ensure that imputation results faithfully reproduce the distribution of observed data. Engineering case study demonstrates that the proposed method achieves superior imputation accuracy with high robustness to missing rates. Moreover, the imputed data significantly improve the performance of downstream tasks, including deformation prediction and anomaly detection, providing a reliable data foundation for comprehensive dam safety assessment.</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

Imputation of missing data for dam deformation based on bidirectional spatiotemporal generative adversarial network

  • 期刊论文

  • 发表日期：

    2025年9月24日

查看全文
  • Hongrui Zhang,
  • Xin Cao,
  • Jinbao Sheng,
  • Chao Jiang,
  • Anjun Zu

摘要

Due to anomalies that occur during data collection, transmission, or storage process in dam health monitoring, deformation data are often incomplete. Accurate imputation of missing deformation data is crucial for assessing dam safety status. While deformation data exhibit spatiotemporal correlations, most existing imputation methods fail to adequately integrate these correlational patterns. To address this issue, this paper proposes a missing deformation data imputation method based on bidirectional spatiotemporal imputation. This method first extracts forward and backward temporal dependency using recurrent neural network, and then integrates spatial relationships among adjacent monitoring points with multilayer perceptron. These imputation results are subsequently combined through an adaptive mechanism. Furthermore, the proposed method is trained within a generative adversarial framework to ensure that imputation results faithfully reproduce the distribution of observed data. Engineering case study demonstrates that the proposed method achieves superior imputation accuracy with high robustness to missing rates. Moreover, the imputed data significantly improve the performance of downstream tasks, including deformation prediction and anomaly detection, providing a reliable data foundation for comprehensive dam safety assessment.