Scholarlink

The deformation analysis of high and steep slopes is crucial for the safety of large-scale hydraulic construction projects. However, existing deformation prediction analysis methods are not sensitive to sudden changes in monitoring data for high and steep slopes, and cannot directly generate safety analysis reports from prediction results. Therefore, this article proposes an innovative method that integrates xLSTM Attention time series prediction and large language models, which not only improves prediction accuracy, but also directly generates structured and engineering reference value safety analysis reports. The experimental results show that the xLSTM Attention model outperforms other benchmark models in time series prediction, with a MAE of 0.163 and an RMSE of 0.238; The performance of the fine tuned large language model in generating evaluation reports is also significant, with BLEU-4 at 16.71%, ROUGE-1 at 36.42%, ROUGE-2 at 13.26%, and ROUGE-L at 21.51%. This method provides an efficient and intelligent solution for slope deformation monitoring and landslide warning in large-scale water conservancy construction projects.

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

BDT-LLM: Beidou-Based Time-Series Prediction and Safety Analysis for High-Steep Slope Deformation Using xLSTM-Attention and Large Language Models

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Lianxing Hu,
  • Qingshan Yang,
  • Xiantao Xu,
  • Li Li,
  • Jianjun Huang,
  • Jirong Tian,
  • Zhaowei Chu,
  • Youchan Huang

摘要

The deformation analysis of high and steep slopes is crucial for the safety of large-scale hydraulic construction projects. However, existing deformation prediction analysis methods are not sensitive to sudden changes in monitoring data for high and steep slopes, and cannot directly generate safety analysis reports from prediction results. Therefore, this article proposes an innovative method that integrates xLSTM Attention time series prediction and large language models, which not only improves prediction accuracy, but also directly generates structured and engineering reference value safety analysis reports. The experimental results show that the xLSTM Attention model outperforms other benchmark models in time series prediction, with a MAE of 0.163 and an RMSE of 0.238; The performance of the fine tuned large language model in generating evaluation reports is also significant, with BLEU-4 at 16.71%, ROUGE-1 at 36.42%, ROUGE-2 at 13.26%, and ROUGE-L at 21.51%. This method provides an efficient and intelligent solution for slope deformation monitoring and landslide warning in large-scale water conservancy construction projects.