The tunnel structure undergoes gradual changes in stress and deformation characteristics over time due to various complex factors during construction and operation. To address this, a deep learning probabilistic model incorporating an attention mechanism is proposed to accurately predict and assess the safety status at critical adverse locations in the tunnel lining. Initially, the Spearman rank correlation coefficient is employed for data preprocessing to select soil pressure and concrete strain data, which are highly correlated with the most adverse locations in the lining structure, as input features. Subsequently, a multi-layer convolutional neural network (CNN) is designed for multi-source data feature extraction, and a feature-sharing layer is constructed to integrate data information from different locations. The extracted features are then fed into a long short-term memory (LSTM) network for time-series analysis and prediction, with an attention mechanism introduced to optimize feature weighting, thereby further improving prediction accuracy. Finally, a Gaussian probabilistic regression model is established to address the quantification and evaluation of uncertainty in safety factor calculations due to prediction errors in structural response. Based on data from a real tunnel engineering project, the response prediction results for adverse locations indicate that the model can comprehensively account for the spatio-temporal correlation of multi-source measurement data. The average prediction errors for concrete strain on the training, validation, and test sets are 0.89, 1.02, and 1.24 , respectively, with no overfitting observed, verifying the proposed method's strong generalization capability in handling complex nonlinear problems. Additionally, a Gaussian probabilistic interval prediction approach is adopted, and a 90% confidence interval estimate is performed using the predicted safety factors. The results show that the safety factors for three critical locations in the secondary lining fall within this confidence interval, further validating the reliability and practicality of the proposed model in assessing the safety of tunnel lining structures.
式中:为l维向量真实值;为预测估计值;为平均值;为向量 y 中第个变量;mean()为求均值的操作。MAE和MSE这2个指标越小越好,R2通常情况下是介于0和1之间的数,越大表示拟合效果越好。一般而言,如果训练集和验证集的指标表现很好,而预测集的指标表现很差的情况,表明模型过拟合,泛化性不够,需要重新训练模型。
HONGK R, FENGH H.Development trends and views of highway tunnels in china over the past decade[J].China Journal of Highway and Transport,2020,33(12):62-76.
WENGB, ZHANGC, QUEY,et al.Mechanical response analysis of tunnel composite pavement based on response surface[J].Forest Engineering,2024,40(2):176-187.
[5]
LAIJ X, QIUJ L, FANH B,et al.Structural safety assessment of existing multiarch tunnel: a case study[J].Advances in Materials Science and Engineering,2017:1-11.
LIUQ S, BAIS Y, XIAOC X,et al.Study on surrounding rock stability of Longtan tunnel based on in-situ monitoring measurements[J].Chinese Journal of Rock Mechanics and Engineering,2007,26(10):1982-1990.
YANGC X, CUIG Y, WANGL M.Shared ratio of pressure in secondary lining based on field monitoring and measuring in highway tunnel[J].World Geology,2012,31(1):193-198.
CHENX S, XUZ H, BAOX H,et al.Challenges and technological breakthroughs in tunnel construction in China[J].China Journal of Highway and Transport,2020,33(12):1-14.
LIM, CHENW Z, YANGJ P.An analysis method for the online monitoring data of tunnel structure[J].Rock and Soil Mechanics,2016,37(4):1208-1216.
[16]
CHENR P, ZHANGP, WUH N,et al.Prediction of shield tunneling-induced ground settlement using machine learning techniques[J].Frontiers of Structural and Civil Engineering,2019,13(6):1363-1378.
ZHAOS M.Study of large deformation prediction of soft rock tunnel based on progressive composite combined prediction model[J].Tunnel Construction,2018,38(7):1131-1137.
YAOK, ZHUX Y, ZHANGK H,et al.Prediction model of surrounding rock deformation in soft rock tunnel based on FOA-GRNN[J].Chinese Journal of Underground Space and Engineering,2019,15(2):908-913.
HUT T, QIANX B, TUP,et al.Prediction of large deformation catastrophes of the carbonaceous slate tunnel in high ground stress based on the optimized support vector machine[J].Journal of Lanzhou University(Natural Sciences),2022,58(5):607-615.
LIUS H, WUC H, CHENX F,et al.Internal force and displacement of composite lining of deep buried tunnels based on load-structure method[J].Journal of Railway Engineering Society,2023,40(11):70-74.
WANGW Z, LIANGQ G, JIAG Y,et al.Statistical analysis of the contact pressure between the primary support and the secondary lining of tunnel[J].Chinese Journal of Underground Space and Engineering,2021,17(5):1586-1597.
YANGJ P, CHENW Z, LIM,et al.Structural health monitoring and response analysis of an underwater shield tunnel during operation[J].Chinese Journal of Rock Mechanics and Engineering,2021,40(5):902-915.
[33]
ZHANGP, WUH N, CHENR P,et al.Hybrid meta-heuristic and machine learning algorithms for tunneling induced settlement prediction: A comparative study[J].Tunnelling and Underground Space Technology,2020,99:103383.
[34]
RAI A, SHRIVASTAVAA, JANAK.A CNN-BiLSTM based deep learning model for mid-term solar radiation prediction[J].International Transactions on Electrical Energy Systems,2021,31(9):e12664.
[35]
VASWANIA, SHAZEERN, PARMARN,et al.Attention is all you need[C].Proceedings of the 31st International Conference on Neural Information Processing Systems.Long Beach,California,USA: ACM,2017:6000-6010.
[36]
LUW, LIJ, WANGJ,et al.A CNN-BiLSTM-AM method for stock price prediction[J].Neural Computing and Applications,2021,33(10):4741-4753.