| With the development of the new western region,more and more traffic projects are advancing to the dangerous mountainous areas,and the tunnel project is recommended as the optimal scheme.In the area of soft rock stratum with high in-situ stress,tunnel excavation will disturb the rock mass,destroy the initial stress state of surrounding rock,and cause the stress redistribution of surrounding rock.With the release of stress,the surrounding rock deforms greatly,resulting in safety accidents such as instability,collapse,support deformation and destruction.If the deformation of surrounding rock can be predicted accurately during tunnel construction,corresponding measures can be taken in advance to prevent accidents and reduce losses caused by construction disasters.Therefore,this thesis uses the theoretical analysis,numerical simulation and field monitoring method,based on the deep learning method,collecting relevant data mining on construction site,the disturbance caused by tunnel construction the deformation prediction technology,development of surrounding rock deformation prediction system,can improve the deformation of soft rock tunnel construction control technology,has the important value of engineering application.The main research contents of this thesis are as follows:(1)The construction deformation mechanism of surrounding rock of tunnel is analyzed from two perspectives: the space-time effect of tunnel construction disturbance and the mechanical mechanism of soft rock tunnel construction.The influencing factors of surrounding rock deformation are divided into internal inducement and external inducement,and the original input characteristics of surrounding rock deformation prediction method are determined.(2)Two deep learning models,Long Short-Term Memory Neural Network and BackPropagation Neural Network,were established to predict surrounding rock deformation during tunnel construction.The results show that: Compared with BPNN,LSTM performs better in deformation prediction,with MAE reduced by 6.3% and 8.1%,and RMSE reduced by 19.2%and 21.4%,respectively,under single feature input mode(input surrounding rock deformation)and multi-feature input mode(input deformation,mechanics and geometrical characteristics).(3)In view of the complexity of tunnel surrounding rock deformation prediction in multifeature input mode,the Convolutional Neural Network is used to optimize the feature extraction ability of the model,and the Temporal Pattern Attention is introduced to give different weights to the input features of hidden states in the Gated Recurrent Unit Network.The model can extract more important input characteristic information from different time steps,and the PSO is used to optimize the model’s hyperparameters,and a tunnel surrounding rock deformation prediction method based on CNN-TPA-GRU is proposed.Compared with GRU model and CNN-GRU model,the MAE of CNN-TPA-GRU model is reduced by 31.5%and 14.2%,and the RMSE is reduced by 38.2% and 21.9%.(4)Based on the tunnel surrounding rock deformation prediction model and migration learning method proposed in this thesis,a set of soft rock tunnel surrounding rock deformation prediction system is developed,and its validity is verified by application in a practical soft rock tunnel project.Comparing the prediction results of the prediction system,numerical simulation results and construction site monitoring data,The results show that: Under the multi-feature input mode(applied in Chenglan Railway tunnel project),the single-step prediction accuracy of the system is the highest,and the MAPE is only 1.13%.The MAPE of the multi-step prediction results of the system is 12.4%,which is better than the numerical simulation results(35.3%).Under the single-feature input mode(applied in Jiumian highway tunnel project),the MAPE of single-step prediction results of the system is 1.33%.When the prediction period is 3 days,the MAPE of multi-step prediction results is 2.9%,and the accuracy can meet the requirements of engineering design. |
PDF Full Text Request