基于图神经网络与时序模型的隧道拱顶沉降时空耦合预测方法OA

Tunnel deformation during construction is jointly influenced by temporal evolution and longitudinal spatial interactions among monitoring sections,exhibiting pronounced nonlinearity and spatiotemporal coupling.To address the limitation of existing methods that predominantly model single-section time series and thus fail to capture nonlinear interactions and spatial effects between adjacent sections,this study proposes a nonlinear spatiotemporal prediction approach for tunnel deformation based on inter-section spatial correlation.A graph neural network(GNN)is used to transform engineering factors,intersection distance,groundwater conditions,and burial-depth differences into spatial correlation weights,enabling adaptive weighted aggregation of static features from neighboring sections.In parallel,a long short-term memory(LSTM)network is employed to capture long-term temporal dependencies in historical deformation records,and is fused with deformation regression for forecasting.The proposed method is validated using in situ monitoring data collected during the construction of the Moxi Tunnel.Results indicate that the method achieves excellent performance in crown settlement prediction on the test set,with a mean absolute error(MAE)of 2.47 mm,a root mean square error(RMSE)of 3.79 mm,and a coefficient of determination(R2)of 0.978.Compared with LSTM,CNN,and CNN-LSTM baselines,the proposed approach delivers substantially improved accuracy and stronger stability,providing reliable technical support for safety risk assessment during tunnel construction.