Random Forest Method-Based Prediction And Control Of Bridge Pier Displacements During Construction Of Shield Tunnels In Close Proximity

With the continuous expansion of underground rail transit construction scale and the continuous improvement of underground space utilization rate in urban areas,the environment for tunnel construction is becoming more and more complex.Among them,the tunnel adjacent to the existing construction,construction of structures is becoming more and more common,tunnel design and construction are facing more severe challenges.Soil pressure balance shield machine(EPBM)has the characteristics of fast construction speed,small disturbance and low risk,and has been widely used in subway tunnel construction in soft soil environment.During the tunnel construction by shield tunneling method,the disturbance of surrounding soil and adjacent buildings and structures is significantly affected by the construction parameters of shield tunneling machine(such as total thrust,palm surface pressure,etc.).Therefore,it is of great significance to determine and set reasonable shield construction parameters to reduce the environmental impact during construction and ensure the safe construction of the project.In recent years,as a new artificial intelligence method,Random Forest method has been widely used in civil engineering.Many studies have shown that the stochastic forest method has the ability to analyze and predict highly nonlinear problems and can assist the construction of geotechnical engineering.In this paper,the vertical displacement of bridge piers during the construction of overlapped shield tunnel of Tianjin metro line 5 is predicted and controlled by random forest method.In this project,the tunnel passes through 78 pier foundations continuously.The pile foundation of four piers(hereinafter referred to as"key piers")is less than 0.5 times the diameter of the tunnel from the net distance of the planned tunnel.However,limited by surface traffic,environment and other conditions,it is impossible to strengthen the pier foundation in advance.In this case,determining and setting reasonable shield construction parameters will be the key to successfully control the vertical displacement of these key piers.In this paper,the random forest method is used to optimize the construction parameters of shield tunneling through the key piers,and the displacement of the four key piers is finally controlled within the specified range.Specifically,it includes the following main work:(1)Machine learning was carried out with the measured data during shield construction of Tianjin metro line 5 overlapped tunnel as the training samples.The random forest model was established and verified.(2)By using the method of random forest to select characteristics,the importance of 10 influencing factors which may cause vertical displacement of pier is ranked,and the predominate variables which should be paid attention to in the construction of upper and lower tunnels are obtained.(3)The vertical displacement of four key piers under different shield construction parameters was predicted by using the random forest model.Then,the Bayesian principle is introduced for statistical analysis,and the main construction variables are optimized.The measured data show that the optimized construction parameters are reasonable and the vertical displacement of the pier can be effectively controlled.(4)After the construction is completed,the measured data of the second group are compared with the predicted results of the random forest model.Two parameters RMSE and R~2 are introduced to evaluate the prediction results.The results show that the random forest model established in this paper can meet the precision requirements of engineering prediction.