Research On Monitoring And Prediction Of A Deep Foundation Pit Construction Deformation In Shenzhen

With the construction of cities,the area available for use in our cities is gradually decreasing.The depth of the pit deepens with the height of the superstructure,and the increased depth of the pit causes that the complexity of the construction process and the importance of the safety requirements are greatly increased.The reduction of available land causes most of the new buildings to be built in the city center,surrounded by pipelines,buildings（structures）and roads.The deformation generated by the pit construction will cause damage to the facilities around the pit,so it is crucial to monitor and predict the deformation of the pit and the surrounding facilities.This paper summarises the current state of research in areas such as the impact of deep foundation pit construction,inverse analysis of geotechnical parameters,numerical simulation of deep foundation pits and machine learning based prediction of pit deformation.A deep foundation pit project in Shenzhen as a research background,the laws and characteristics of construction deformation of the research project are derived from the analysis of actual monitoring data.A three-dimensional numerical simulation and LSTM model were used to predict the construction deformation of deep foundation pits,the conclusions are as follows:（1）Deformation prediction of foundation pits is mostly using numerical simulation or machine learning prediction.Numerical simulation can predict the distribution pattern and change trend of deformation generated by foundation excavation.However,the ground and the enclosure structure are still slowly deforming after the excavation of the foundation pit is finished,and the numerical simulation cannot make better prediction for the construction stage after the excavation of the foundation pit due to the limitations of parameters and models,etc.Machine learning methods can predict the trend of individual points more accurately,and can also effectively predict the construction phase after the end of the pit excavation,but they require a certain amount of data as support.This paper combines the advantages of both numerical simulation and machine learning methods to predict the deformation of the enclosure structure,surrounding ground settlement and pipeline deformation throughout the construction process.（2）The temporal pattern of deformation from pit excavation increases as the pit construction progresses.The spatial pattern is that the horizontal displacement of the enclosure structure,ground settlement and pipeline deformation are distributed in a pattern of large on the central axis of the pit and small on both sides.The horizontal displacement deformation law of the deep layer of the enclosure structure is the largest at the top.The horizontal displacement deformation of the enclosure structure at depth is maximum at the top and decreases with increasing depth.The Surface settlement settles first and then decreases away from the enclosure.The law of foundation pit deformation depends on the support form and excavation sequence,and the deformation distribution law summarized in this paper is only applicable to the foundation pit targeted in this study.（3）Numerical simulation calculations require input of geotechnical and material mechanical parameters.The samples of the indoor geotechnical tests are not entirely consistent with the actual state of the geotechnical samples,and the parameters obtained after the experiments may deviate from the actual geotechnical parameters.The use of geotechnical parameter inversion can reduce the effects of disturbance.This paper uses particle swarm combined with multioutput least-squares support vector regression machine to invert geotechnical parameters The results of the inversion parameters are generally better than those of the experimental parameters and are closer to the actual deformation.Numerical simulations with inversion of parameters allow for effective prediction of deformations in foundation pit projects.（4）In this paper,the deformation of the construction phase after the end of excavation of the foundation pit is predicted by using the multi-measurement point LSTM model,and the prediction results have less error compared with the monitoring data,which can provide some reference for other projects.In order to verify the superiority and practicality of the multi-measurement point LSTM,two sets of comparison experiments are conducted.A group of results comparing LSTM model,time series model and BP neural network for pit deformation prediction.The results show that the LSTM model has higher accuracy in predicting the construction deformation of foundation pit and can effectively reflect the trend of construction deformation of deep foundation pit.The other group is to compare the prediction effect of multi-point LSTM over single-point LSTM,and the results show that multipoint LSTM has better prediction effect.