Scholarlink

<p>Structural health monitoring (SHM) systems play a critical role in ensuring the safety of buildings during seismic events. However, their effectiveness is limited when sensor data is lost due to malfunction, damage, or communication failure. To address this issue, this study proposes a bi-directional urban safety network that utilizes LSTM models to predict the dynamic structural responses of buildings based on the responses of adjacent structures within the network. The framework incorporates both time-domain displacement data and frequency-domain features, which are transformed into conditional vectors for input into the LSTM model. The proposed method was evaluated using both linear and nonlinear structural systems subjected to seismic loads. Results demonstrate that the use of conditional vectors significantly improves prediction accuracy, with RMSE reductions of up to 27.13% compared to baseline models. Additionally, the method accurately predicts maximum response amplitudes and time-dependent nonlinear behavior. These findings suggest that the proposed framework offers a robust and scalable approach for structural response recovery in sensor-deficient urban environments, enhancing seismic resilience at a regional scale.</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

LSTM-based bi-directional urban safety network using a conditional vector of frequency domain decomposition data

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年8月25日

查看全文
  • Han Yong Lee,
  • Insub Choi,
  • Byung Kwan Oh

摘要

Structural health monitoring (SHM) systems play a critical role in ensuring the safety of buildings during seismic events. However, their effectiveness is limited when sensor data is lost due to malfunction, damage, or communication failure. To address this issue, this study proposes a bi-directional urban safety network that utilizes LSTM models to predict the dynamic structural responses of buildings based on the responses of adjacent structures within the network. The framework incorporates both time-domain displacement data and frequency-domain features, which are transformed into conditional vectors for input into the LSTM model. The proposed method was evaluated using both linear and nonlinear structural systems subjected to seismic loads. Results demonstrate that the use of conditional vectors significantly improves prediction accuracy, with RMSE reductions of up to 27.13% compared to baseline models. Additionally, the method accurately predicts maximum response amplitudes and time-dependent nonlinear behavior. These findings suggest that the proposed framework offers a robust and scalable approach for structural response recovery in sensor-deficient urban environments, enhancing seismic resilience at a regional scale.