关键词:

寒区隧道, 冻结深度, 正交试验, XGBoost模型, LightGBM模型, 混合模型

Abstract:

The onsite measured data of the Xing′anling tunnel and the numerical simulation results are studied together to investigate the characteristics of the tunnels′ freezing depth in cold regions and further predict the freezing depth. Therefore, the temperature field and the distribution law of the freezing depth are explored. Subsequently, based on the orthogonal test, the sensitivity and significance of various influencing factors of the freezing depth are analyzed. Additionally, a hybrid prediction model is proposed based on extreme gradient boosting and light gradient boosting machine models to predict the freezing depth. The results reveal the following: (1) As the distance from the tunnel portal increases, the freezing depth of the surrounding rock shortens; moreover, the freezing depth of the surrounding rock varies sharply when close to the tunnel portal, whereas the freezing depth tends to be stable when the distance from tunnel portal is larger than 300 m. (2) The freezing depth′s change rule in different locations of each crosssection is similar. There is an obvious increase in the freezing depth in the inverted arch and no obvious difference in other parts. (3) The sensitivities of the freezing depth to various factors are ranked as follows: initial ground temperature>average temperature in the coldest month>specific heat capacity of the surrounding rock>thermal conductivity of the surrounding rock>thermal conductivity of the lining>specific heat capacity of the lining. (4) The proposed hybrid prediction model based on XGBoost and LightGBM model has a higher accuracy and stronger applicability than the traditional singlemodel prediction method.

Key words:

tunnel in cold region, freezing depth, orthogonal test, extreme gradient boosting, light gradient boosting machine, hybrid model