Scholarlink

<p>The early stages of large landslide evolution often leave detectable traces, and understanding the development process and deformation mechanisms during this creep phase is crucial for effective landslide prevention and control. However, due to the complexity and long duration of landslide formation, capturing the initial evolution of landslides at a low cost remains a significant challenge in current research. In this study, the Mogrifier algorithm is integrated into the Long Short-Term Memory (LSTM) framework, resulting in the M-LSTM algorithm. The Luoduzhai potential landslide was selected as the study area. Two algorithms were used to learn historical deformation data for predicting future deformation, and the superiority of the M-LSTM algorithm was verified through comparison with actual values and the RMSE index. M-LSTM overcomes the limitations of previous methods in which information interacted only within the gate. Using the Small Baseline Subset Interferometric Synthetic Aperture Radar (SBAS-InSAR) technology, nine years of accumulative deformation data are obtained for potential landslides. By combining this data with multi-dimensional matrices derived from various geological factors, we train the model to predict future deformation trends and mechanisms of potential landslides, thereby achieving a more precise understanding of their future evolution process. Comparative results show that the improved M-LSTM algorithm significantly outperforms LSTM, with 79.18% of deformation point errors within 10&#xa0;mm for M-LSTM while LSTM is 75.55%. Combined with field investigations, Luoduzhai potential landslide were analyzed to be a push-type landslide. Its failure mode involves the upper part driving the lower part, ultimately leading to an overall slide. The evolutionary trend indicates stability around the slope edges and continuous deformation within the slope body. These findings provide valuable guidance for future landslide disaster prevention and mitigation efforts.</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

A multi-dimensional modified deep learning framework for predicting potential creep-type landslides based on geological data

  • 期刊论文

  • 发表日期：

    2026年5月23日

查看全文
  • Yue Liu,
  • Peihua Xu,
  • Chen Cao,
  • Liling Jin,
  • Bo Han,
  • Rongchen Ma,
  • Chensi Li

摘要

The early stages of large landslide evolution often leave detectable traces, and understanding the development process and deformation mechanisms during this creep phase is crucial for effective landslide prevention and control. However, due to the complexity and long duration of landslide formation, capturing the initial evolution of landslides at a low cost remains a significant challenge in current research. In this study, the Mogrifier algorithm is integrated into the Long Short-Term Memory (LSTM) framework, resulting in the M-LSTM algorithm. The Luoduzhai potential landslide was selected as the study area. Two algorithms were used to learn historical deformation data for predicting future deformation, and the superiority of the M-LSTM algorithm was verified through comparison with actual values and the RMSE index. M-LSTM overcomes the limitations of previous methods in which information interacted only within the gate. Using the Small Baseline Subset Interferometric Synthetic Aperture Radar (SBAS-InSAR) technology, nine years of accumulative deformation data are obtained for potential landslides. By combining this data with multi-dimensional matrices derived from various geological factors, we train the model to predict future deformation trends and mechanisms of potential landslides, thereby achieving a more precise understanding of their future evolution process. Comparative results show that the improved M-LSTM algorithm significantly outperforms LSTM, with 79.18% of deformation point errors within 10 mm for M-LSTM while LSTM is 75.55%. Combined with field investigations, Luoduzhai potential landslide were analyzed to be a push-type landslide. Its failure mode involves the upper part driving the lower part, ultimately leading to an overall slide. The evolutionary trend indicates stability around the slope edges and continuous deformation within the slope body. These findings provide valuable guidance for future landslide disaster prevention and mitigation efforts.