Prediction And Control Of Tunneling-induced Settlement Using Machine Learning Algorithms

Earth pressure balanced shield tunneling inevitably disturbs the surrounding ground,leading to decrease of the performance and structural safety of buildings and structures around the tunnel.The degree of disturbance is affected by tunnel geometric parameters,geological parameters and shield operational parameters,but the existing theory and models are difficult to comprehensively consider the interaction of these factors and accurately predict the ground response.In order to solve the these problems,this study uses machine learning algorithm to establish the prediction model of tunneling-induced settlement characteristics,which provides a new idea for real-time prediction of tunneling-induced ground response and reducing risk.The main results of this study are summarized as follows:(1)A database including surface settlement and influencing factors was established.Based on the field data,influencing factors consider tunnel geometric parameters,operational parameters,geological parameters and anomalous conditions,and the output variable considers the maximum ground surface settlement and the settlement trough width.A new method to quantify the geological parameters was proposed.Based on the evaluation of the effects of different influential parameters on the settlement,the parameters with significant correlation on the surface settlement are selected as the input parameters.The results indicate that the proposed geologic parameter quantification method can comprehensively consider ground physical and mechanical properties,and the geometrical characteristics of soils depth and thickness.The geometric parameters(tunnel depth),operational parameters(thrust,torque,chamber pressure,penetration rate,grouting ratio),geological parameters(water table,ground conditions at the tunnel face,modified count of standard penetration test,modified count of dynamic penetration test,modified uniaxial compressive strength),and anomalous shield stoppage,are significant on the settlement.These 12 parameters are thus used as the input parameters of the machine learning model to predict the settlement.(2)The performances of different machine learning algorithms in the prediction of tunneling-induced settlement were compared.Based on selected performance evaluation indicators,mean absolute error(MAE),root mean square error(RMSE)and coefficient of determination,R~2,the performance of six machine learning algorithms(BP neural network,wavelet neural network,generalized regression neural network,extreme learning machine,support vector machine,and random forest)in predicting tunneling-induced settlement is compared comprehensively.K-fold cross-validation method is employed to determine the optimal hyper-parameters of each machine algorithm.The results indicate that the prediction results of BP neural network vary dramatically,losing fidelity at some monitoring points.Wavelet neural network,generalized regression neural network,extreme learning machine and support vector machine can not accurately predict the evolution of settlement.The prediction results of the random forest algorithm show great agreement with the measured settlement,and it can accurately capture the evolution of settlement.(3)Prediction of settlement trough and recognizing anomalous settlement condition based on random forest(RF)algorithm was conducted.Optimal machine learning algorithm RF was used to establish the prediction models of maximum settlement and settlement trough width.The integration with Peck formula can be used to predict settlement trough.The shape of the lateral settlement tank caused by the excavation.Meanwhile,RF algorithm was also used to recognize tunneling-induced settlement condition(whether or not exceed the warning value 10 mm).The results indicate that the RF-based models can accurately predict the settlement and recognize settlement condition.(4)Hybrid heuristic optimization algorithm particle swarm optimization and random forest algorithm was proposed to establish an intelligent control model.A database consisting of reasonable set of operational parameters(ie,the corresponding settlement is less than 10 mm)was establish,which was employed to establish a shield operational parameters prediction model.By inputting geological parameters and tunnel geometry parameters,this model can estimate the reasonable operational parameters at the design stage.When the suggested value of the operational parameter cannot control the settlement under 10 mm,the hybrid algorithm further optimizes the operational parameters to achieve the control of surface settlement.The results indicate that the tunneling-induced settlement based on the predicted operational parameters is much smaller than the settlement induced by the measured operational parameters.The proposed hybrid algorithm can control the shield tunneling process in a real time,intelligent and effective way.