Research On Prediction Method Of Land Subsidence Based On Deep Learning

Shield construction will cause soil compression,resulting in ground settlement.When the local surface settlement is too large,it will seriously affect the safety of the construction personnel and cause significant economic losses.Therefore,the ground settlement during the shield construction process is accurately predicted in advance.And early warning has important practical significance.The current methods of surface settlement prediction have the following problems: the data collected during construction are not fully utilized;a large number of experiments are needed to analyze the law of surface settlement;the real-time prediction of surface settlement cannot be achieved;In view of the above problems,this paper is proposed a method for predicting surface subsidence based on deep learning.The main research contents are as follows:(1)Construct a general research framework for prediction of surface subsidence.Based on the study of the deformation law of the surface subsidence and the analysis of the mechanism of surface subsidence,the overall research framework for the prediction of surface subsidence is proposed.The main contents are as follows: first,data preprocessing,feature engineering,then constructing and optimizing a model for predicting surface subsidence and settlement trend,and finally defining targets for surface subsidence early warning and setting reasonable safety thresholds.(2)Screening of key influencing factors of surface subsidence.In view of the complex shield construction environment and many influencing factors,feature engineering was used to screen out the key factors that have a greater impact on ground settlement.The model of surface subsidence selection based on the random forest and distance correlation coefficient method is constructed separately,and the key factors that have a large impact on surface subsidence are screened in combination with the mechanism knowledge.(3)Constructing a prediction model of ground surface subsidence based on the optimization of DNN neural network by the search algorithm of beetles.Aiming at the characteristics of large amount of shield construction data and high dimensions,a prediction model of ground surface settlement based on DNN neural network was proposed,and the standard beetles search algorithm was improved.The improved initial weight of the DNN neural network model was improved by using the improved beetles search algorithm,which improved the model's convergence speed and the prediction accuracy of the model.(4)Constructing a prediction model of ground surface subsidence based on LSTM neural network optimized by the Monochamus alternatus search algorithm and early warning of surface subsidence.Aiming at the problems of large surface subsidence and long duration in the construction of dangerous areas,a prediction model of surface subsidence based on DNN neural network was proposed,and the initial weight of DNN neural network model was optimized by using the improved search algorithm of beetles.On the basis of the predicted settlement amount and settlement trend,the target definition of the ground settlement is set,a reasonable safety threshold is set,and the danger information is alarmed in advance.Based on the above research content and using the construction data of Xi'an Metro Line 5,a case analysis of surface settlement prediction was completed,which verified the effectiveness of the method proposed in this paper.