Scholarlink

<p>Seismic monitoring remains a significant challenge in geophysical research due to the complex and dynamic nature of seismic precursor signals. In this study, we introduce a novel anomaly detection framework (F-VAE-RL), which integrates a frequency-aware Variational Autoencoder (VAE) with reinforcement learning to identify irregularities in borehole strain data. The F-VAE-RL framework is specifically designed to capture both long-term and short-term frequency patterns, enabling effective differentiation of subtle signals from background noise in multivariate time series. By leveraging these latent representations, the reinforcement learning module actively explores high-risk segments and adaptively refines the detection process. Experimental evaluations on field-collected datasets demonstrate that F-VAE-RL outperforms traditional reconstruction-based methods, achieving superior precision in detecting pre-seismic anomalies. By combining frequency-domain analysis with active decision-making, F-VAE-RL provides a robust and adaptive solution for seismic monitoring, significantly enhancing the safety and reliability of detection systems.</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

Frequency-aware variational autoencoder with reinforcement learning for borehole strain anomaly detection

  • 期刊论文

  • 发表日期：

    2025年10月10日

查看全文
  • XiaoLong Wei,
  • QingJie Liu,
  • ZhiAn Pan

摘要

Seismic monitoring remains a significant challenge in geophysical research due to the complex and dynamic nature of seismic precursor signals. In this study, we introduce a novel anomaly detection framework (F-VAE-RL), which integrates a frequency-aware Variational Autoencoder (VAE) with reinforcement learning to identify irregularities in borehole strain data. The F-VAE-RL framework is specifically designed to capture both long-term and short-term frequency patterns, enabling effective differentiation of subtle signals from background noise in multivariate time series. By leveraging these latent representations, the reinforcement learning module actively explores high-risk segments and adaptively refines the detection process. Experimental evaluations on field-collected datasets demonstrate that F-VAE-RL outperforms traditional reconstruction-based methods, achieving superior precision in detecting pre-seismic anomalies. By combining frequency-domain analysis with active decision-making, F-VAE-RL provides a robust and adaptive solution for seismic monitoring, significantly enhancing the safety and reliability of detection systems.