Research On Shield Axial Deviation Prediction Based On LSTM And Correction Method

With the rapid development of society and economy and the continuous acceleration of urbanization,the demand for subway,highway and railway tunnel construction is getting higher and higher.Due to its safety,high efficiency and environmental protection,shield tunneling has become the mainstream construction method for tunnels.During the shield construction process,due to the complex geological environment,line alignment and other factors,the axis of the shield machine is likely to deviate,and the deviation of the axis deviation is difficult to predict.When the deviation exceeds the allowable range,the parameters of the shield machine during the correction process,the setting is difficult to determine,and the correction depends on manual experience.When it deviates from the threshold,if it is not corrected in time,it will cause serious safety hazards.Therefore,this paper takes the shield construction axis as the research object,proposes a method for predicting shield axis deviation based on LSTM neural network,and proposes a multi-loop correction method.The main research contents are as follows:(1)Construction of a shield axis deviation prediction and correction analysis framework.Combed the shield construction process,and analyzed the business principle of axis deviation and excavation correction in shield construction,and then based on the data mining analysis method to propose the axis deviation prediction and correction method.It mainly includes: data preprocessing,analysis of influencing factors,construction of prediction model of axis deviation,and construction of multi-loop correction model.(2)Analysis of the important influence factors of axis deviation.Shield construction data includes shield tunneling data,shield attitude data,grouting data,and status data.Due to the large amount of data,complex attributes,high dimensions,and low data quality and knowledge density,direct modeling cannot guarantee the model Accuracy and timeliness.After data cleaning,this paper combines machine learning methods with statistical analysis for feature selection,and combines the results of mechanism analysis to reconstruct the features of the data to construct an analysis data package,which lays the foundation for subsequent model training efficiency and accuracy.(3)Construction of axis deviation prediction model based on LSTM network.During the shield tunneling process,the axis deviation is controlled by the tunneling parameters and is changed based on the deviation from the previous moment.Based on the many factors affecting the axis deviation and the sequence of the deviation data,Using the state information memory ability and good sequence data processing ability of the LSTM network,an axial deviation prediction model based on LSTM was proposed,at the same time,Dropout was added to optimize the model to improve the prediction accuracy of the model and the generalization ability of the model.(4)Construction of a shield axis correction model.Aiming at the problem of shield construction,in order to ensure that the shield tail clearance is within a safe range and avoid potential safety hazards,this paper proposes a correction model combining geometric calculation and data mining technology.Mainly include: First,through geometric calculation,obtain the total number of loops required for correction and the correction amount of each loop;Secondly,establish a regression model for the variation of the deviation,and obtain the functional relationship between the variation of the deviation and the parameters of the shield tunneling;Then the data is discretized,and the association rules between the deviation variation interval and the tunneling parameter interval are mined;then the geometric calculation results are matched with the association analysis interval to obtain the tunneling parameter interval corresponding to each loop deviation;Finally,the regression model is used to iterate in the interval,and a set of data with the smallest absolute loss is taken as the optimal parameter recommended value.This method guides the parameter setting by mining the information contained in the historical correction data,avoiding the limitation of artificial experience.Based on the above research content,take a company’s shield project as an example for modeling analysis.The results show that this method has achieved ideal results in the prediction and correction of the deviation of the shield construction axis,which verifies the effectiveness of the method.