Scholarlink

An in-depth analysis of Structural Health Monitoring (SHM) data is essential for ensuring structural safety, with anomaly detection playing a crucial role in this assessment. This study optimizes an anomaly detection algorithm using modern machine learning techniques for static monitoring data, including ambient temperature, humidity, and strain, specifically in Traditional Tibetan Timber Structures (TTTSs). The research begins with a comprehensive analysis of the characteristics of the monitoring data, followed by the establishment of a thorough data preprocessing process. This process employs the sigma-LOF algorithm to identify anomalies and the NeuralProphet algorithm to address missing data. Building on this foundation, we propose an LSTM-MHSA-VAE neural network that integrates a Long Short-Term Memory Network (LSTM), a Multi-Head Self Attention (MHSA) mechanism, and a Variational autoencoder (VAE) with a state threshold criterion to effectively detect and diagnose anomalies in the monitoring data. The results demonstrate that this model significantly improves accuracy in identifying and diagnosing anomalies, surpassing the effectiveness of traditional methods.

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

Anomaly Detection of Static Monitoring Data of Traditional Tibetan Timber Structures Based on LSTM-MHSA-VAE Algorithm

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Xiaowen Qian,
  • Na Yang

摘要

An in-depth analysis of Structural Health Monitoring (SHM) data is essential for ensuring structural safety, with anomaly detection playing a crucial role in this assessment. This study optimizes an anomaly detection algorithm using modern machine learning techniques for static monitoring data, including ambient temperature, humidity, and strain, specifically in Traditional Tibetan Timber Structures (TTTSs). The research begins with a comprehensive analysis of the characteristics of the monitoring data, followed by the establishment of a thorough data preprocessing process. This process employs the sigma-LOF algorithm to identify anomalies and the NeuralProphet algorithm to address missing data. Building on this foundation, we propose an LSTM-MHSA-VAE neural network that integrates a Long Short-Term Memory Network (LSTM), a Multi-Head Self Attention (MHSA) mechanism, and a Variational autoencoder (VAE) with a state threshold criterion to effectively detect and diagnose anomalies in the monitoring data. The results demonstrate that this model significantly improves accuracy in identifying and diagnosing anomalies, surpassing the effectiveness of traditional methods.