In-situ Data Modeling Methods Of Shield Machine And Their Application Research

With the advent of the data-intensive times,the operational monitoring of the shield machines is becoming increasingly perfect.The measured in-situ data not only record the important information of the operation process but also involve the internal mechanism of the shield machine and the internal mechanism with the working environment.It is of great significance in improving the design,analysis,operation,and maintenance level of the shield machine through mining these in-situ data.The in-situ data of the shield machine usually consists of the operation in-situ data from the monitoring system and the geological in-situ data from the geological investigation.It has the following features:the operation in-situ data are overlapped and are not labeled with multiple geological conditions,the relationship between the attributes of the operation in-situ data and the performance of the shield machine is complex,the collected operation in-situ data can not cover the sample space of subsequent data,and the sample size of the operation in-situ data is not equal to that of the geological in-situ data.In view of the above features of the in-situ data modeling methods of the shield machine,this dissertation carried out the following four aspects of research:1)the partition method for the operation in-situ data of the shield machine;2)the accurate modeling method for the operation in-situ data of the shield machine;3)the generalization ability improvement method for the prediction model of the operation in-situ data of the shield machine;4)the modeling method for the operation and geological in-situ data of the shield machine.The advance rate is the key performance index of the shield machine,which describes the tunneling distance or tunneling efficiency per unit time,and is the essential reference for adjusting operation parameters and project construction management.Based on the in-situ data from a tunnel project of Shenzhen metro,the feasibility of the proposed models and methods is verified through the experiments of the advance rate prediction.The main contributions are as follows:(1)For the problem of partitioning the operation in-situ data of different working conditions,an attribute correlation-guided clustering method is proposed.A linear equation is used to describe the attribute correlation and incorporated into the clustering objective function of the proposed method.The objective function optimization strategy of interactive iteration of membership,prototypes,and linear equation coefficients is designed to obtain the optimum partition results.The clustering experiments of the operation in-situ data of a shield machine show that the proposed method can effectively partition the operation in-situ data of different working conditions and provide the regression equations of the correlated attributes.(2)For the problem of accurate modeling of operation in-situ data of the shield machine,a clustering model based on regression relationship is proposed,based on which a data clustering-assisted data modeling method is proposed.In the proposed clustering model,a clustering objective function is constructed based on the prediction errors of the regression model of each cluster,and an algorithm is designed to optimize it to obtain the clustering results.Base on the data clustering information,a classification model is built to classify the data to be predicted.And,this data is inputted into the corresponding regression model to obtain its output.From the advance rate prediction results based on operation in-situ data,it is found the proposed data modeling method can accurately evaluate the relationship between the operation parameters and the eadvance rate,and provide more conpetitive prediction results compared with the other data modeling methods.(3)In the operation in-situ data of the shield machine,the collected data often cannot effectively cover the space of the subsequent data,which leads to the problem of insufficient generalization ability of the prediction model.The simulation data is introduced to improve the generalization ability of the operation in-situ data.Since the simulation data parameters of the shield machine are not the same as the operation in-situ data,a data modeling method for the multi-fidelity data with different parameters is proposed.A new kernel function is designed to map the input of multi-fidelity data to high-dimensional feature space,and the regression relationship between the input and output of multi-fidelity data is evaluated by a linear equation in the high-dimensional feature space.The experimental results of advance rate prediction of shield machine indicate that the proposed data modeling method can take advantage of the simulation data and operation in-situ data to improve the prediction accuracy and generalization ability of the advance rate prediction model.(4)For the problem of data modeling of the operation and geological in-situ data,a data fusion method is proposed to converts the numerical data of the geological distribution and categorical data of geological classification data into new numerical data firstly.Then,a data modeling method is proposed for the data of unequal sample sizes,and used for modeling the operation in-situ data and geological in-situ data of the shield machine.A new kernel function is designed to map the inputs of the data of larger sample size and the data of smaller sample size into high-dimensional feature space,and then a linear model is used to evaluate the relationship between inputs and outputs in the high-dimensional feature space.The proposed method is applied to build the advance rate prediction model of a shield machine based on the geological in-situ data of smaller sample size and the operation in-situ data of larger sample size.It is found that the proposed method can accurately predict the dynamic and cumulative advance rate.The effect of the geological and operation parameters on the advance rate is also studied.Finally,the main works of this dissertation are summarized,and the future works for the in-situ data modeling methods of the shield machine have prospected.