Scholarlink

<p>To address the rainfall effect on slope stability, this study presents a novel hybrid architecture that integrates long short-term memory (LSTM) networks, interpolation techniques, and convolutional neural networks (CNNs). This architecture uses LSTMs to predict pore pressure (POR) at sparse monitoring points over time, uses interpolation algorithms to distribute these predictions across the slope domain, and uses CNNs to assess slope stability. This integrated approach mitigates the limitations of existing methods for forecasting temporal variations in pore water pressure from sparse data and effectively incorporates the derived spatial distribution into stability analysis. First, historical POR data from monitoring points are used to train the LSTM model to predict future POR values. The predicted values are then extrapolated across the slope using an interpolation algorithm. Integrating this spatially distributed POR with slope geometry and material properties enables the CNN to predict the factor of safety (FS). The results demonstrate that the root mean square error (RMSE) for the artificial intelligence (AI)-predicted factor of safety (FS) is 0.02 on the testing dataset and 0.03 for the actual slope case, relative to numerical simulation benchmarks. Furthermore, the predicted factor of safety exhibits an initial decrease followed by an increase, which closely corresponds with variations in rainfall intensity. This correlation confirms the efficacy of the proposed method in predicting slope stability dynamics under fluctuating rainfall conditions. Thus, it provides a valuable foundation for developing enhanced landslide early warning 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

Time series prediction of the slope stability under rainfall conditions based on LSTM and CNN

  • 期刊论文

  • 发表日期：

    2025年10月3日

查看全文
  • Mansheng Lin,
  • Yucheng Lu,
  • Yan Li,
  • Gongfa Chen,
  • Bingxiang Yuan

摘要

To address the rainfall effect on slope stability, this study presents a novel hybrid architecture that integrates long short-term memory (LSTM) networks, interpolation techniques, and convolutional neural networks (CNNs). This architecture uses LSTMs to predict pore pressure (POR) at sparse monitoring points over time, uses interpolation algorithms to distribute these predictions across the slope domain, and uses CNNs to assess slope stability. This integrated approach mitigates the limitations of existing methods for forecasting temporal variations in pore water pressure from sparse data and effectively incorporates the derived spatial distribution into stability analysis. First, historical POR data from monitoring points are used to train the LSTM model to predict future POR values. The predicted values are then extrapolated across the slope using an interpolation algorithm. Integrating this spatially distributed POR with slope geometry and material properties enables the CNN to predict the factor of safety (FS). The results demonstrate that the root mean square error (RMSE) for the artificial intelligence (AI)-predicted factor of safety (FS) is 0.02 on the testing dataset and 0.03 for the actual slope case, relative to numerical simulation benchmarks. Furthermore, the predicted factor of safety exhibits an initial decrease followed by an increase, which closely corresponds with variations in rainfall intensity. This correlation confirms the efficacy of the proposed method in predicting slope stability dynamics under fluctuating rainfall conditions. Thus, it provides a valuable foundation for developing enhanced landslide early warning systems.