Deformation monitoring of large-scale complex structures is crucial for operational safety. Traditional methods face challenges in capturing complex nonlinear deformations and integrating multi-source heterogeneous data. This paper proposes a physically motivated CNN-Transformer hybrid deep learning model, which employs a dual-branch parallel architecture to capture both the local nonlinear features induced by stress concentration in the deformation field and the global long-range spatiotemporal dependencies arising from the coordinated deformation of the overall structure. A multi-source feature fusion module based on an attention mechanism is also designed to achieve synergistic complementarity between ground-based sensors and InSAR data at the feature level. Taking the main stadium of Xi'an Olympic Sports Center as a case study, experiments are conducted using monitoring data during its construction and operation periods. The results show that the CNN-Transformer model achieves a root mean square error (RMSE) of 0.82 mm, a mean absolute error (MAE) of 0.63 mm, and a coefficient of determination (R²) of 0.981 for deformation monitoring, significantly outperforming comparison models such as measurement adjustment, SBAS-InSAR, standalone CNN/Transformer, and CNN-LSTM at a statistically significant level (p < 0.01). This study provides an effective paradigm for intelligent and high-precision deformation monitoring of large-scale infrastructure.
ZHUAnlong, ZiqianDA, LIUJianyou, et al. Intelligent monitoring of structural safety of badaling Great Wall Station on Beijing-Zhangjiakou High-speed Railway[J]. 2020, 64(1): 94-98.
LIUYanwei, PENGJie, RENLianwei, et al. Horizontal deformation prediction of retaining piles for deep foundation pit of subway station based on CEEMDAN-SSA-ELM-LSTM Model [J]. Journal of Disaster Prevention and Mitigation Engineering, 2025, 45(1): 34-46.
YUHanyang, LANChaozhen, WANGLonghao, et al. Research progress and trends in deep learning for multi-modal remote sensing image matching [J]. Journal of Geo-information Science, 2025, 27(08): 1896-1919.
SONGJintao, ANChence, YANGJie, et al. A deformation prediction method for hydraulic engineering slopes based on the fusion of GNSS and deep learning [J]. Journal of Water Resources and Water Engineering, 2025, 36(1): 129-137.
GAOFuqiang, LIUWenju. Laser scanning and deep learning-based measurement methods and techniques for forces of rock bolt groups[J]. Journal of China Coal Society,2026,51(2):994-1006..
YANGXin, PANYanliang, CHENHuijiang, et al. Application analysis of 3D laser scanning in disease treatment and deformation monitoring of subway structures[J]. Geomatics & Spatial Information Technology, 2025 (9): 175-177.
LIUGe, DENGXingsheng. Building detection from remote sensing images combining deep learning and graph cuts [J]. Bulletin of Surveying and Mapping, 2019 (11): 69-73.
LIULing, WANGGuofeng, LIJiancheng, et al. Building extraction method of airborne laser scanning based on object analysis strategy[J]. Journal of Chang' an University (Natural Science Edition), 2015, 35(S1):135-139.
ZhumingOU, JIANGLule, XIEJian, et al. A deformation prediction model for deep foundation pits based on transfer learning [J]. Journal of Zhejiang University of Technology, 2025, 53(4): 392-398.
SUXiaojun, XUZengguang, ZHANGYe, et al. Research on concrete dam deformation prediction based on deep learning model [J]. Yellow River, 2025, 47(7):144-149,155.
HUANGYu, LIUKankan, CHENGTianxiao, et al. Intelligent prediction model for deformation induced by excavation of harbor foundation pit based on temporal fusion transformer[J]. Journal of Basic Science and Engineering, 2025, 33(5): 1287-1296.
ZHONGQiaojun, ZHAOErfeng, HULingzhi, et al. A deep learning prediction model for ultra-high arch dam deformation based on ACNN-VAR [J]. Water Resources and Power, 2025, 43(4): 163-167.
YITingjingwen, HUANGCaisheng, QINYong, et al. Transformer-CNN prediction model of high and steep slope deformation based on Bedidou detection data[J]. Journal of Chongqing University, 2025,48(10):81-94.
ZHANTao, HUANGHao. Research on settlement prediction of adjacent buildings based on shield construction process and space-time factors[J]. China Civil Engineering Journal, 2024,57(S1):117-121.
AIXingxing, LIUXingyang, QIUJianchun, et al. A deep learning model for concrete dam deformation considering dynamic spatiotemporal correlation Characteristics [J]. China Rural Water and Hydropower, 2025, (12): 47-54.
LIUHongbo, YANGZhifeng, ZHOUTing,et al. Intelligent drawing review of steel frame structure joints based on revit secondary development and deep learning[J]. Journal of Building Structures, 2024,45(7):43-55.
XIANGZhenyang, BAOTengfei, BAIYanli, et al. A dam deformation prediction model based on hybrid attention mechanism and deep learning [J]. Advances in Science and Technology of Water Resources, 2023, 43(2): 96-101.
CAIAn, ZHENGDongjian. Intelligent prediction model for dam deformation based on adaptive decomposition and temporal feature fusion[J]. South-to-North Water Transfers and Water Science & Technology,2026,24(1):261-272.
ZHANGYuming, WANGEndong. Comparative analysis of dam deformation monitoring models based on interpretability [J]. Water Conservancy Science and Technology and Economy, 2025, 31(8): 79-83,90.
LIUHuanchen, ZHUJing, GUOMengjing. A deep learning prediction model for arch dam deformation integrating characteristic factor screening [J]. Water Resources and Hydropower Engineering, 2025, 56(3): 123-134.
GAOTao. Hybrid prediction model for slope deformation by integrating gaussian-filter and deepAR[J/OL].Construction Technology, (2024-5-16)[2025-10-31].