Study On Prediction And Parameter Configuration Of Ground Subsidence In Shield Construction Based On Deep Learning

Accurately and efficiently predict surface settlement,the advancing speed and grouting amount of tunnel shield tunneling are the keys to safe and fast excavation of tunnels.However,how to optimize equipment parameters to ensure that tunnels are excavated at the expected advancing speed.There is no better method or uniform standard.At present,the construction personnel basically rely on experience to set various parameters,so they cannot efficiently control the shield machine to work in the desired state.Based on this engineering problem,this article models and analyzes the data collected on the construction site.The main tasks are as follows:(1)This paper takes the control engineering of the interval tunnel under the high-speed bridge as the background,with the help of numerical simulation software and the controlled variable method,comprehensively analyzes the law of surface deformation caused by the changes of geological parameters and excavation parameters during shield tunnel excavation.Several factors are most sensitive to surface settlement.With the help of finite element analysis results,the original data collected by the shield machine was preliminarily screened and denoised,and then dimensionality was reduced based on the pca method and Pcc method.On the premise of retaining 95% of the effective features of the original data,the original data was successfully reduced.The data has been reduced from 12 dimensions to 7 dimensions.(2)Based on the pre-processed data set,two LSTM neural network models and traditional BP neural network models are used,and geological parameters,geometric parameters and shield machine parameters are used as independent variable inputs to maximize the ground caused by the tunnel construction process.Settlement is predicted.The results show that the coefficient of fit of the multi-factor LSTM model reaches 0.917,and the relative error does not exceed 0.01,which is an excellent completion of the prediction of surface settlement;the single-factor LSTM model and BP neural network also meet the performance requirements of the project.(3)In order to ensure the tunneling of the shield machine at the expected grouting volume and advancing speed,with the aid of the settlement prediction model framework,a prediction model of the advancing speed and grouting volume of the LSTM neural network is established.Among them,the determination coefficient of fit of the propulsion speed prediction model reached 0.961,and the determination coefficient of fit of the grouting volume prediction model reached 0.947.The rm seand the aae did not exceed 0.2,which made the prediction excellent.Based on the results,it is assumed that the allowable ground settlement at 3 meters in front of the shield machine is 2mm,and the recommended values of the advancing speed and synchronous grouting parameters are given.In summary,the use of neural network methods for ground settlement prediction and shield parameter analysis is to keep up with the general trend of intelligent development,combine new theories such as big data analysis with actual engineering,and promote traditional shield construction towards Moving forward in a safer and more efficient direction.