Research And Application Of Treatment Methonds For Incomplete Data Of Deformation Monitoring In Foundation Pits

It is necessary to perform deformation monitoring on the engineering body during the construction phase and the operation phase,in order to ensure the normal operation of the foundation pit project.The key to the deformation monitoring of foundation pits is the processing of data.The purpose is to predict deformation.At present,commonly used data processing methods include single models such as multiple linear regression,gray model,time series model and combined models between two or more single models,but these methods are all based on the premise of full data.Due to human error,data may be missing or unavailable in the process of acquisition of the foundation pit monitoring data,resulting in incomplete data.The commonly used incomplete data processing methods are deletion,padding and Kalman filtering.These methods improve the quality of data processing to certain extent,but its limitations of missing data filling,makes the processing precision of incomplete data is limited by filling quality of data.This paper attempts to introduce the theories and methods of mature incomplete data processing in other fields,so that these theories and methods can serve the deformation monitoring work.Therefore,based on the expectation-maximization(EM)algorithm in the field of mathematical statistics and the parameter estimation in the modeling of autoregressive AR(p)model,this paper takes the data obtained from the monitoring of the foundation pit as the research object.The treatment method using EM algorithm is discussed when there is incomplete data in the foundation pit deformation monitoring.In this paper,the application of EM algorithm in incomplete deformation monitoring data is studied as follows:(1)Analyze the data processing theory and methods under incomplete data through the commonly used data processing methods of the foundation pit deformation monitoring in the field of measurement,and discuss the applicable conditions,advantages and disadvantages of various methods.(2)According to the principle and extension of the EM algorithm and its applicability in Gaussian normal distribution,combined with the parameter estimation in the AR(p)model modeling process,the analysis of the EM algorithm in comparison with the data incomplete Least squares has the advantage.(3)Combining with specific engineering examples,the application of EM algorithm in the parameter estimation of AR(p)model,in contrast to deletion and filling in the application of incomplete deformation monitoring data processing,and verify the feasibility of EM algorithm.The superiority of EM algorithm in the case of single missing data and double missing data.The applicability of the EM algorithm is verified by comparing the prediction results with the GM-AR model and BP neural network model.