Scholarlink

<p>Digital twin technology has emerged as a promising management tool for the construction and operation of hydraulic engineering. However, timely parameter synchronization between excavation-induced changes and numerical models remains challenging due to high computational costs and varying construction conditions. If the timeliness and accuracy of parameter updating are not guaranteed, stability analysis and security assessment will be delayed and inaccurate. This study proposes an adaptive model updating framework for parameter synchronization of reservoir slope digital twin systems, based on Gaussian process regression, adaptive optimization, and Pareto-guided active learning. The Gaussian process regression is employed to identify spatial heterogeneous deformation from monitoring data. The adaptive optimization links&#xa0;the multi-stage excavation process&#xa0;with Pareto-guided active learning, and&#xa0;the latter selectively augments the training dataset by focusing on high-impact regions of the parameter space. This framework is validated on the excavation of the left bank slope of the Batang (BT) hydropower station in southwest China. Results demonstrate that the approach accelerates model updating by approximately three to four times compared with the conventional method, with the mean absolute error of the displacement fitting results being less than 1%. Moreover, an unstable signal near the spillway tunnel exit was detected prior to the observed collapse. This study presents an efficient and reliable solution for model updating during excavation, supporting timely stability assessment and enhancing the synchronization capability of reservoir slope digital twin 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

Adaptive Excavation Slope Model Updating Framework Accelerated by Pareto-Guided Active Learning

  • 期刊论文

  • 发表日期：

    2026年5月19日

查看全文
  • Yanji Chen,
  • Di Wang,
  • Huan Zhao,
  • Qianru Ding,
  • Xiang Cheng,
  • Wei Zhou,
  • Gang Ma

摘要

Digital twin technology has emerged as a promising management tool for the construction and operation of hydraulic engineering. However, timely parameter synchronization between excavation-induced changes and numerical models remains challenging due to high computational costs and varying construction conditions. If the timeliness and accuracy of parameter updating are not guaranteed, stability analysis and security assessment will be delayed and inaccurate. This study proposes an adaptive model updating framework for parameter synchronization of reservoir slope digital twin systems, based on Gaussian process regression, adaptive optimization, and Pareto-guided active learning. The Gaussian process regression is employed to identify spatial heterogeneous deformation from monitoring data. The adaptive optimization links the multi-stage excavation process with Pareto-guided active learning, and the latter selectively augments the training dataset by focusing on high-impact regions of the parameter space. This framework is validated on the excavation of the left bank slope of the Batang (BT) hydropower station in southwest China. Results demonstrate that the approach accelerates model updating by approximately three to four times compared with the conventional method, with the mean absolute error of the displacement fitting results being less than 1%. Moreover, an unstable signal near the spillway tunnel exit was detected prior to the observed collapse. This study presents an efficient and reliable solution for model updating during excavation, supporting timely stability assessment and enhancing the synchronization capability of reservoir slope digital twin systems.