Research On TBM Construction Schedule Simulation Of Long Distance Diversion Tunnel Based On Improved XGBoost Parameter Updating

Tunnel Boring Machine(TBM)construction is an important way of long distance diversion Tunnel construction at present.However,the complex construction conditions and changeable construction environment of TBM bring some difficulties to the accurate analysis and management of construction progress.Construction simulation technology is a powerful means for accurate analysis of construction progress.Due to the influence of site construction environment,construction resources and other factors,TBM tunneling time,step-change time,support time and other construction parameters are not invariable,but will constantly change with the progress of construction.Therefore,it is necessary to dynamically update the construction simulation parameters according to the actual construction situation,so as to accurately track the construction status and improve the accuracy of simulation results.The existing simulation parameter update based on Bayesian update requires the assumption of prior distribution,and the parameter distribution form remains unchanged in the update process,which is inconsistent with the actual situation,and it is difficult to accurately describe the nonlinear variation of construction simulation parameters.The machine learning method has a good effect on nonlinear parameter prediction.EXtreme Gradient Boosting(XGBoost)is more accurate and easier to tune than shallow machine learning.Therefore,this paper established the improved XGBoost model for TBM construction simulation parameter updating of diversion tunnel,and proposed the TBM construction simulation method based on the improved XGBoost parameter update,and achieved the following main research results:(1)An improved XGBoost model for dynamic updating of TBM construction simulation parameters was established.The artificial jellyfish multi-objective optimization algorithm was used to optimize the four hyperparameters of XGBoost model,namely,learning rate,optimal number of iterations,maximum optimization depth and random seed,so as to reduce the workload of manual parameter tuning and improve the model accuracy and modeling efficiency.Secondly,in order to reduce the subjectivity of time window selection in time series prediction of construction parameters,partial autocorrelation function(PACF)is used to determine the prediction time window according to the correlation between data.(2)According to the discrete event simulation principle,a TBM construction simulation method based on improved XGBoost parameter updating is proposed by using object-oriented graphic-assisted modeling method and hierarchical modeling method,combined with the actual construction characteristics of long distance diversion tunnel TBM.(3)The method proposed in this paper is applied to the TBM construction schedule simulation of a diversion tunnel,and the excavation time is taken as an example to update and predict the construction parameters.The improved XGBoost parameter update results are compared with the results based on bayesian update method、 without improved XGBoost.The results show that the proposed model has higher prediction accuracy of construction parameters(mean absolute error is 19.66,mean square error is 659.39,mean percentage error is 25.99%,and correlation coefficient is 0.96),which further ensures the accuracy of simulation results.