Study on Multi-Factor Coupling Analysis of Temperature Field and Freezing Radius Prediction for Cold-Region Tunnels

To clarify the evolution laws of surrounding-rock temperature field and freezing radius in cold-region tunnels under the combined action of natural wind and train piston wind, and to improve the capability of rapid frost-damage risk assessment, a joint study of numerical simulation and machine learning was conducted using the Guolashan Tunnel as an engineering background. Based on field parameters, a three-dimensional transient heat transfer model consisting of air, primary support, secondary lining, and surrounding rock was established. Non-isothermal flow was used to couple the airflow with the heat transfer process of the lining and surrounding rock, and an equivalent wind speed method was adopted to characterize the effects of train piston wind and residual wind. Seven types of factors, including natural wind speed, natural wind temperature, natural wind direction, tunnel cross-section size, initial rock temperature, train speed, and train operation frequency, were analyzed through the control variable method. The longitudinal and radial temperature distributions and the freezing radius after 90 days were extracted, and random forest, support vector machine, and XGBoost prediction models were established based on 658 sets of orthogonal samples. The results indicate that the temperature-adjustment zone at the tunnel portal expands with the increase of cold-air action time; the longitudinal temperature of the surrounding rock presents two-stage characteristics of rapid rise and stable growth, and the radial temperature gradually approaches the initial rock temperature. Natural wind temperature, natural wind direction, and initial rock temperature are the dominant factors, with sensitivity coefficients of 0.22, 0.21, and 0.21, respectively. The XGBoost model has the optimal prediction accuracy with a root mean square error of 0.118; the predicted freezing radius for the engineering case is 1.85 m, with a relative error of 1.1%. External low-temperature air conditions and the initial thermal state of the surrounding rock are the key factors for frost-damage risk control in cold-region tunnels. The established models can provide a basis for rapid estimation of the freezing radius and optimization of anti-freezing design parameters.