CNN-GRU-BASED DEFORMATION PREDICTION METHOD FOR SUBWAY EXCAVATIONS
Keywords:
Foundation pit, Deformation prediction, CNN, GRUAbstract
To address the issues of decreased deformation prediction accuracy and early warning lag caused by missing monitoring data during subway foundation pit construction, this paper proposes a deformation prediction method based on the CNN-GRU neural network. First, a convolutional neural network (CNN) is used to impute missing data in the original monitoring sequence. Second, wavelet decomposition is applied to decompose the imputed sequence into a trend component and a noise component. Third, a gated recurrent unit (GRU) neural network is constructed to predict the trend component, while an ARMA model is used to fit and predict the noise component. Finally, the prediction results of the two parts are reconstructed to obtain the final deformation prediction value. This paper conducts an empirical analysis using the vertical displacement monitoring data of the wall top of No. 2 air vent at Terminal 3 Station of B Airport. The results show that with a missing rate of 5.8%, the proposed model achieves a mean absolute percentage error (MAPE) of 1.93% and a coefficient of determination (R²) of 0.95. Compared with GA-BP, LSTM, single GRU, and ARMA models, all error metrics are significantly reduced, and the prediction lag problem is effectively improved. This method provides an effective solution for deformation prediction of foundation pits under incomplete data conditions.References
[1] Liu Mengyang. Research on deformation prediction and active control of deep foundation pit excavation in subway stations. Huazhong University of Science and Technology, 2024. DOI: 10.27157/d.cnki.ghzku.2024. 002630.
[2] Zhou Yi, Wang Zhangqiong, Zou Yuangeng, et al. Research on Prediction Method of Metro Foundation Pit Deformation Based on CNN-GRU Neural Network with Incomplete Data. Urban Rail Transit Research, 2025, 28(06): 32-36+53. DOI: 10.16037/j.1007-869x.20230612.
[3] Liu Cheng, Liu Yong, Rong Wenzhen, et al. Research on Prediction of Deformation of Foundation Pit Based on CNN-Attention-LSTM//Process Control Professional Committee of Chinese Society of Automation, Chinese Society of Automation. Proceedings of the 35th China Conference on Process Control. School of Automation and Information Engineering, Sichuan University of Light Chemical Technology; Sichuan Provincial Key Laboratory of Artificial Intelligence, 2024: 879. DOI: 10.26914/c.cnkihy.2024.020078.
[4] Yao Ronghan, Zhang Wensong, Jia Lei, et al. Settlement estimation model of buildings around subway foundation pit based on spatiotemporal characteristics and deep learning. Journal of Rock Mechanics and Engineering, 2025, 44(S1): 196-205.
[5] Gao Y, Xiao Z, Gong Z, et al. Spatiotemporal Deformation Prediction Model for Retaining Structures Integrating ConvGRU and Cross-Attention Mechanism. Buildings, 2025, 15(14): 2537.
[6] Cao Jing, Ding Wenyun, Zhao Dangshu, et al. Prediction of foundation pit deformation time series based on LSSVM-ARMA model. Rock and Soil Mechanics, 2014, 35(S2): 579-586. DOI: 10.16285/j.rsm.2014.s2.021.
[7] Bao Yanni, Shen Danyi, Shi Zhenming, et al. Application of ARMA model in the prediction of deformation of anchor foundation pits. Journal of Engineering Geology, 2021, 29(05): 1621-1631. DOI: 10.13544/j.cnki.jeg. 2021-0527.
[8] Ma Zhigang, Yang Guolin, Liu Tao, et al. Wavelet GRU-ARMA Optimized InSAR Monitoring Settlement Prediction Method. Surveying and Mapping Bulletin, 2024, (12): 33-39. DOI: 10.13474/j.cnki.11-2246.2024.1206.
[9] Fang Qing, Chen Sheng, Liu Xuezhu, et al. Application of CNN-LSTM model based on variational mode decomposition in foundation pit deformation prediction. Mechanics and Practice, 2024, 46(05): 1015-1022.
[10] Tang Haoran, Hu Yao, Lei Huayang, et al. Prediction of surface displacement caused by foundation pit excavation based on deep learning. Transactions of the Chinese Society of Geotechnical Engineering, 2024, 46(S2): 236-241.
[11] Shi Xiangfeng, Wang Lifen, Shen Yang, et al. Research on Time Series Prediction of Ground Settlement in Foundation Pit Construction Based on GA-SVM. Construction Technology, 2017, 46(08): 16-19+31.
[12] Yang Zhongxiang, Zhang Liangyu, Lai Xingxiang, et al. Prediction of deformation of CNN-PSO-LSTM deep foundation pit considering spatial feature fusion. Shanxi Architecture, 2026, 52(07): 108-114+138. DOI: 10.13719/j.cnki.1009-6825.2026.07.021.
[13] Ma Zhigang. Extraction of surface settlement characteristics and optimization prediction of wavelet GRU-ARMA in mining area based on InSAR technology. Lanzhou Jiaotong University, 2024. DOI: 10.27205/d.cnki.gltec. 2024.001540.
[14] Li Sihui, Liu Haiqing. Prediction of deformation of deep foundation pit based on LMD-PSO-LSSVM combined model. Journal of Underground Space and Engineering, 2018, 14(02): 483-489. DOI: 10.20174/j.juse.2018.02. 026.
[15] Wang X, Qin Z, Bai X, et al. Research Progress of Machine Learning in Deep Foundation Pit Deformation Prediction. Buildings, 2025, 15(6): 852.
[16] Hu Dong, Zhang Xiaoping. Research on Prediction of Foundation Pit Deformation Based on Gray System Theory. Journal of Underground Space and Engineering, 2009, 5(01): 74-78+168.
[17] Han Feng. Research on combined prediction of foundation pit deformation under data filtering conditions. Geotechnical Engineering Technology, 2023, 37(04): 392-396.
[18] Liang Xiaolong, Qi Erheng, Wang Qiangkun, et al. Research on Combined Prediction Model of Foundation Pit Deformation Wavelet Denoising and BP Neural Network. Surveying, Mapping and Spatial Geographic Information, 2021, 44(01): 189-192+195.