Research On Deformation Prediction Method Of Deep Foundation Pit Based On Gaussian Process Regression Model

With the rapid development of the economy,the pace of infrastructure construction in China is accelerating,and deep foundation pits are often formed during the construction of underground infrastructure.Deformation monitoring of deep foundation pit retaining structures,modeling based on monitoring data and predicting future deformation values is an important task to prevent disasters such as foundation pit collapse.There are many existing modeling and prediction methods,but for the deformation of nonlinear and high-dimensional numbers,the current prediction modeling methods are also flawed and less accurate.Based on the superiority of Gaussian process regression theory in modeling prediction,this paper studies the deep foundation pit deformation prediction modeling method based on Gaussian process regression theory.The specific research contents are as follows:(1)Simulation experiments studied the influence of a single covariance function on the prediction accuracy of the Gaussian process regression model.Because the Gaussian process regression model has good generalization ability,it has good fitting ability to the training samples in most cases.For the qualitative and quantitative results,the prediction effects of five commonly used single covariance functions on modeling are used.The function performs Gaussian process regression modeling and compares the prediction accuracy of the model.The prediction accuracy of the model is better when the neural network(NN)covariance function is used in the simulation.(2)A preferred experiment for a single covariance function.Select five commonly used single covariance functions: square index(SE),Maten(Matern32,Matern52),neural network(NN),period(PER),linear(LIN)function,respectively construct a Gaussian process regression model,study The model fitting data and deformation prediction ability,and comparing the analysis results,the optimal single covariance function is the Matern Matern32 function.(3)A preferred experiment for combining covariance functions.According to the optimal single covariance function,Matthew Matern32 adopts the additive combination form,and forms five combined covariance functions with SE,Matern52,NN,PER and LIN,and constructs a Gaussian process regression model to study the model fitting data and deformation.Predictive ability,comparative analysis results show that the overall performance of the model of the combined covariance function is higher than the model of the single covariance function,and the optimal combined covariance function is the Matern32+Matern52 combination function.(4)Construct a dynamic prediction model of the optimal training sample under the number of periods.Under the premise of selecting the optimal combination covariance function(Matern32+Matern52),the model fitting and prediction accuracy comparison experiments of training samples with different periods are carried out.The experimental results show the dynamics of the training samples when the logarithmic value is 14 The predictive model performs optimally.(5)Research on dynamic prediction model under different driving factors.The correlation analysis between the deformation value of deep foundation pit and time,historical deformation data and adjacent point deformation values shows strong correlation.Based on the optimal training sample logarithm and optimal combination covariance function,three dynamic prediction models are constructed: 1 "time" single-factor driven Gaussian process regression dynamic prediction model,2 "time + historical deformation data" two-factor driving combination Gaussian process regression dynamic prediction model,3 "time + historical deformation data + adjacent point deformation value" three-factor driven combined Gaussian process regression dynamic prediction model.The comparison of the accuracy of the prediction results shows that the three-factor-driven combined Gaussian process regression model has the best overall performance.The three-factor-driven Gaussian process regression dynamic prediction model based on "time + historical deformation data + adjacent point deformation value" proposed in this study is more accurate than the traditional prediction model,and the output of the model can be set with certain confidence.Interval,using probabilistic meaning to express the credibility of the predicted value,is more suitable for the deformation prediction of some large-scale high-risk projects,providing more scientific and effective prediction for the deformation of the monitoring body,thus ensuring the safety construction of the project.