基于时序网络的隧道冻结圈的环境响应预测  

Environmental Response Prediction of Tunnel Freeze Ring Based on Time Series Network

作　　者：张国远 杨兴烁 雷宇旭 于啸波 ZHANG Guoyuan;YANG Xingshuo;LEI Yuxu;YU Xiaobo(Guangxi Key Laboratory of Rock and Soil Mechanics and Engineering,Guilin University of Technology,Guilin Guangxi 541004,China;School of Civil Engineering,Guilin University of Technology,Guilin Guangxi 541004,China)

机构地区：[1]桂林理工大学广西岩土力学与工程重点实验室,广西桂林541004 [2]桂林理工大学土木工程学院,广西桂林541004

出　　处：《兰州工业学院学报》2025年第2期1-6,共6页Journal of Lanzhou Institute of Technology

基　　金：国家自然科学基金资助项目(42067041);广西岩土力学与工程重点实验室课题(桂科能19-Y-21-5);桂林理工大学科研启动基金项目(GUTQDJJ2019043)。

摘　　要：为了快速评估隧道冻结全生长过程中周围环境变化特点,提出时间序列分析模型双向长短时记忆网络(BiLSTM)对冻结过程中的土体的温度场、位移场、孔隙水压力和土压力进行预测。以广州地铁3号线为原型开展室内物理模型试验,将结果的前20%作为训练数据集,后续80%的结果作为测试数据集,对BiLSTM的四组不同的超参数组合进行预测结果对比,得出BiLSTM_n64h2的效果最佳,它的测试集的R^(2)为0.926,RMSE为10.238,MAE为4.87,MAPE为28。通过ABAQUS数值模拟与BiLSTM的对比可发现,BiLSTM拥有更好的预测精度与速度,而ABAQUS由于局限于多个假设会出现较多的误差。In order to quickly evaluate the characteristics of the surrounding environment during the freezing process of the tunnel,a time series analysis model BiLSTM is proposed to predict the temperature field,displacement field,pore water pressure and soil pressure of the soil during the freezing process.The indoor physical model test is carried out with Guangzhou Metro Line 3 as the prototype,and the first 20%of the results are taken as the training data set,and the next 80%are taken as the test data set.By comparing the prediction results of four different groups of BiLSTM hyperparameter combinations,it is found that BiLSTM_n64h2 has the best effect,and its test set R^(2)is 0.926,RMSE is 10.238,MAE is 4.87,and MAPE is 28.By comparing ABAQUS numerical simulation with BiLSTM,it can be found that BiLSTM has better prediction accuracy and speed,while ABAQUS has more errors due to the limitation of multiple assumptions.

关 键 词：冻结法 隧道土体 BiLSTM 温度场 位移场 ABAQUS 

分 类 号：TU973[建筑科学—结构工程]