Geological Recognition And Prediction Of TBM Advance Rate In Tunneling Construction Based On In-Situ Data

Tunnel boring machine is the most widely used heavy load equipment in tunnel construction.Because it mainly relies on the cutter head at the front end of the equipment to continuously rotate and cut the rock and soil media to achieve tunneling,it constitutes a complex engineering system with continuous mechanical interaction with the surrounding geology.With the rapid development of sensor detection technology,the current tunneling equipment can record all kinds of in-situ data information related to the operation status in the construction process,which provides sufficient data source basis for analyzing the interaction rule of key parameters in the tunneling system and predicting its change trend.This kind of engineering in-situ data often has the characteristics of highly coupling and non-linear among the parameters.With the development of intelligent control and decision-making of large engineering systems,how to effectively analyze the engineering in-situ data has become a hot issue in the field of engineering Science In recent years,the rapid development of various machine learning methods provides a powerful tool for the analysis of engineering in-situ data with their excellent non-linear expression ability.Geological recognition and prediction of advance rate are two key problems that are directly related to the efficient and safe construction of tunnel boring machine.This paper first establishes a geological type recognition system based on the analysis of the in-situ data,which combines the chi-square test feature selection method and the machine learning classification algorithm,including the normalization preprocessing,the chi-square test selection to determine the input features and the machine learning method to train the ground quality recognition classifier.Based on the analysis of the properties of the in-situ data,the parameters sensitive to geological change are selected from hundreds of influence parameters by chi square test feature selection method after normalization preprocessing,and then the geological recognition model is established based on four machine learning methods,which are tested in different tunnel engineering.In addition,the generalization ability of the selected features and the geological recognition model is discussed in the test data of different tunnel projectsBased on the recognition of different geology,this paper analyzes the sequence properties of in-situ data,and establishes the prediction model of geological parameters based on the RNN(recurrent neutral network)methods.Furthermore,considering the correlation between different geological parameters,multi-objective regression method is used to predict the geological parameters.The prediction model of geological parameters is tested in Tianjin Metro Line 9,and the prediction performance of different time RNN methods and multi-objective regression methods is discussed.In addition,based on RNN,this paper establishes the prediction model of advance rate,discusses the application characteristics of lazy learning and eager learning methods in the in-situ data,as well as the outlier value existing in the in-situ data,also explores the influence of hyper parameter selection in the machine learning method in the analysis of noisy.The work of this paper shows that through the statistical analysis and machine learning of the measured data of the project,it can effectively describe the influence law among the key parameters such as the load,geological conditions and equipment in the tunneling project,realize the recognition and prediction of some key engineering parameters,and provide a reference for the parameter prediction and intelligent control of a complex engineering system related to the mechanical action.