Scholarlink

<p>Climate warming and engineering activities have exacerbated permafrost degradation in the lower part of the project, leading to significant subgrade deformation in cold regions. To ensure the safe operation of railways, it is necessary to predict subgrade deformation. This study innovatively investigates the deformation prediction of permafrost subgrades from the perspective of road-environment system coupling, transcending the limitations of traditional engineering scales. Using environmental, permafrost, and engineering data along the Qinghai-Tibet Railway collected through aerial, satellite, and ground-based techniques, a spatially distributed dataset covering 25 key conditioning factors (2,660 sample sets) is constructed. This study uses Random Forest (RF) as the base model and integrates two swarm metaheuristic optimization algorithms, that is, Grey Wolf Optimization (GWO) and Particle Swarm Optimization (PSO), while incorporating a K-fold cross-validation mechanism to construct a stepwise optimization paradigm for predicting subgrade deformation. This paradigm progresses step-by-step through four levels: baseline model establishment, intelligent hyperparameter optimization, introduction of cross-validation strategies, and comparison and selection of optimization algorithms. Model evaluation is conducted across prediction accuracy and error, prediction uncertainty, and dimensionless reliability dimensions. The results show that the K-GWO-RF model demonstrated the best predictive performance, followed by the K-PSO-RF and PSO-RF models, with the standard RF model ranking last. This ranking confirms that the optimization strategies at each level yielded quantifiable performance gains. The K-GWO-RF model also significantly outperforms other commonly used models (Layered Summation Method, Backpropagation Neural Networks and Support Vector Machines). Futhermore, using the Mean Decreasing Accuracy calculation for feature importance analysis, it is determined that the main disaster-causing factors of subgrade deformation include solar radiation, thermokarst lakes, rain, and snow depth. This study realizes the accurate prediction of deformation for permafrost subgrades, which provides important technical support and data reference for the rational design, safe operation, and control of deformation diseases of railway projects in cold regions.</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

Study on the predictive model of subgrade deformation of the Qinghai-Tibet Railway in permafrost regions based on multi-source data

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年8月3日

查看全文
  • Qingzhi Wang,
  • Zhi Wen,
  • Zhiwei Zhou

摘要

Climate warming and engineering activities have exacerbated permafrost degradation in the lower part of the project, leading to significant subgrade deformation in cold regions. To ensure the safe operation of railways, it is necessary to predict subgrade deformation. This study innovatively investigates the deformation prediction of permafrost subgrades from the perspective of road-environment system coupling, transcending the limitations of traditional engineering scales. Using environmental, permafrost, and engineering data along the Qinghai-Tibet Railway collected through aerial, satellite, and ground-based techniques, a spatially distributed dataset covering 25 key conditioning factors (2,660 sample sets) is constructed. This study uses Random Forest (RF) as the base model and integrates two swarm metaheuristic optimization algorithms, that is, Grey Wolf Optimization (GWO) and Particle Swarm Optimization (PSO), while incorporating a K-fold cross-validation mechanism to construct a stepwise optimization paradigm for predicting subgrade deformation. This paradigm progresses step-by-step through four levels: baseline model establishment, intelligent hyperparameter optimization, introduction of cross-validation strategies, and comparison and selection of optimization algorithms. Model evaluation is conducted across prediction accuracy and error, prediction uncertainty, and dimensionless reliability dimensions. The results show that the K-GWO-RF model demonstrated the best predictive performance, followed by the K-PSO-RF and PSO-RF models, with the standard RF model ranking last. This ranking confirms that the optimization strategies at each level yielded quantifiable performance gains. The K-GWO-RF model also significantly outperforms other commonly used models (Layered Summation Method, Backpropagation Neural Networks and Support Vector Machines). Futhermore, using the Mean Decreasing Accuracy calculation for feature importance analysis, it is determined that the main disaster-causing factors of subgrade deformation include solar radiation, thermokarst lakes, rain, and snow depth. This study realizes the accurate prediction of deformation for permafrost subgrades, which provides important technical support and data reference for the rational design, safe operation, and control of deformation diseases of railway projects in cold regions.