In the red-bed areas of southwestern China, subgrade uplift deformation poses a serious safety concern for high-speed trains. However, the subgrade uplift mechanisms are still not well-defined, and there is a lack of effective prediction methods for addressing this issue. The objective of this study is to build prediction model of subgrade uplift using three machine learning techniques (MLTs): artificial neural network (ANN), random forest (RF), and support vector machine (SVM). The Chengdu-Chongqing passenger dedicated line (CCPDL) was selected as the research object, and a total of 200 cuttings along the CCPDL were randomly divided into two groups: a training set (70%) and a testing set (30%). The subgrade uplift mechanism was concluded by conducting the laboratory test, field investigation and mathematical statistics. Then six subgrade uplift-conditioning factors were identified, including subgrade excavation height, subgrade excavation width, dip angle, interbedded characteristics between sandstone and mudstone, mudstone rheology, and mudstone swelling. To assess the model performance, various evaluation metrics were employed, including receiver operating characteristic curve (ROC), area under the curve (AUC), accuracy, precision, recall, specificity, and F-1 score. The results demonstrate that the RF model outperforms the other MLTs in predicting subgrade uplift. Notably, among the six factors considered, subgrade excavation height was identified as the most influential factor. These findings provide valuable insights into the prediction of subgrade uplift and offer guidance for mitigating the risks associated with subgrade uplift during the construction of high-speed railways.
Full Text Finder