Research On Deformation Prediction Model Of Deep Foundation Pit In Reclamation Area Based On Machine Learning

In recent years,with the rapid growth of population and the increase of urban underground space development intensity in coastal cities,land resources have become a scarce element of coastal cities.In order to meet the needs of human life and economic development,large amounts of land resources have been obtained through reclamation.Due to the particularity of the soil layer in the reclamation area and the complexity of the surrounding environment,it is easy to produce excessive deformation in the excavation process of deep foundation pit in the reclamation area,and it could endanger the safety of deep foundation pit itself and surrounding environment.Therefore,taking Shenzhen Airport Station of the Shenzhen-Dongguan-Shenzhen Intercity Railroad as the supporting project,which is a typical project in the Shenzhen reclamation area,theoretical analysis,numerical simulation,field monitoring and machine learning and other methods were be comprehensively used to carry out research.Thus,deformation mechanism of deep foundation pit in reclamation area,influence factors of deep foundation pit deformation in reclamation area,prediction model of maximum deformation of deep foundation pit in reclamation area and prediction model of deformation trend of deep foundation pit in reclamation area were researched.The main research results are as follows:(1)The existing research results of deep foundation pit deformation in reclamation area were investigated.It is concluded that the deep foundation pit engineering in reclamation area has the characteristics of weak and complex soil layer,high density of surrounding buildings and strong time-space effect,which leads to excessive deformation in the process of deep foundation pit excavation.It is difficult to predict by traditional methods.(2)The calculation result of deep foundation pit deformation of supporting project shows that the deformation of deep foundation pit increases gradually during excavation,and the distribution form of ground settlement curve changes from triangle to groove.The settlement after last excavation reaches a peak of 12.79 mm at a distance of 6.5m from the edge of the foundation pit,and the influence range is about twice the depth of the foundation pit.With the increase of excavation depth,the distribution form of the horizontal displacement curve of diaphragm wall changes from triangle to parabola,and the position of the peak displacement moves downward.The horizontal displacement of the diaphragm wall after last excavation reaches a peak of 29.25 mm at a distance of 15m(roughly 3/4 of the excavation depth)from the ground surface.(3)The influence law of influencing factors on the maximum deformation of deep foundation pit is studied from the perspectives of soil parameters,design parameters and surrounding overload.The results show that the value of soil parameters has a great impact on the deformation calculation results.The elastic modulus of soil has the greatest influence,followed by the internal friction angle,and finally the cohesion.The deformation of the foundation pit can be effectively controlled by increasing the thickness of the diaphragm wall,the insertion ratio of the diaphragm wall,the concrete strength grade of the diaphragm wall and the section size of the concrete support,as well as reducing the spacing of the steel support.However,the deformation can not be effectively limited by blindly increasing the design parameters.The surrounding overload is closely related to the deformation of deep foundation pit.Reducing overload can better limit the deformation of deep foundation pit.(4)Based on the calculation results of influencing factors of deep foundation pit deformation in reclamation area,the key influencing factors affecting the maximum deformation of deep foundation pit are selected by grey correlation analysis method.The key influencing factors are used as input parameters,and the maximum deformation caused by deep foundation pit excavation is used as output parameters.All calculation conditions and their corresponding maximum deformation are integrated into data sets.According to the ratio of 4:1,the training set and the testing set are divided for learning and testing.An improved particle swarm optimization BP neural network(PSO-BP)deep foundation pit maximum deformation prediction model is constructed.When the improved PSO-BP model is used to predict the maximum ground settlement,the average absolute error is 0.605,the root mean square error is0.685,the maximum absolute error is-1.06 mm,and the maximum relative error is-11.31%.When the improved PSO-BP model is used to predict the maximum horizontal displacement of diaphragm wall,the mean absolute value error is 1.035,the root mean square error is 1.356,the maximum absolute error is-2.72 mm,and the maximum relative error is-11.46%.The model has high prediction accuracy and good stability.The correctness of the model optimization idea is verified by comparing the performance of the model with the ordinary PSO-BP model and BP neural network model on the test set.(5)The actual deformation monitoring data of the supporting project are used as the data set.After preprocessing,the data set is divided into training set and testing set according to the ratio of 7:3 for learning and testing,and the prediction model of deep foundation pit deformation based on genetic algorithm optimized LSTM network(GA-LSTM)is constructed.The test results show that the variation law of the predicted value and the measured value of deep foundation pit deformation are basically consistent.The residual value and error rate fluctuate slightly around 0.The variation range of residual value is [-0.44 mm,0.55mm],and the variation range of error rate is [-6.41%,5.91%].The average absolute error value is 0.238,and the root mean square error value is 0.283.The GA-LSTM deep foundation pit deformation situation prediction model has accurate prediction effect.By comparing the performance of GA-LSTM model with LSTM and BP neural network model on the same test set,it is concluded that the GA-LSTM deep foundation pit deformation situation prediction model performs best.