Research On TBM Tunnel Surrounding Rock Classification Method And Tunnelling Parameters Optimization Based On Machine Learning

At present,China has become the country with the largest number of tunnels under construction and to be built in the world.A considerable part of the tunnels are long tunnels and deep buried tunnels,which are faced with complex geological environment problems and extremely difficult to construct.TBM construction has become the general trend of tunnel construction in the future because of its advantages of high efficiency,good quality,safe operation and friendly environment.However,due to the high sensitivity of TBM to formation conditions and poor adaptability to adverse geological conditions,TBM plays an important role in driving efficiency.At the same time,due to the constraints of many factors such as engineering geological conditions and tunnelling technology level in the construction process of TBM,the dynamic interaction law of rock mechanics parameters is not well understood,and it is difficult to timely and accurately judge the surrounding rock conditions of heading face and predict the tunnelling performance of TBM.In most cases,it depends on the experience of the main driver to make decisions on the tunnelling parameters.In view of the above problems,this paper takes TBM tunnel surrounding rock classification and tunnelling parameter optimization decision as the main research objectives,adopts machine learning,data mining,model test,engineering verification and other means,establishes a real-time identification model of TBM tunnel surrounding rock grade based on the actual tunnelling data of a project in Xinjiang,and carries out cluster analysis on the field tunnelling data,and establishes a set of surrounding rock classification model suitable for TBM tunnel According to the rock classification system,the corresponding prediction model and optimization model are established for the prediction and optimization of tunnelling parameters,and verified by model test.The main research work and achievements of this paper are as follows:(1)Under the geological conditions of gneiss and granite,the relationship between TBM driving parameters and surrounding rock grade is studied.The distribution of thrust,rotational speed,torque,penetration rate,advance rate,utilization rate,penetration,field penetration index and excavation specific energy in all grades of surrounding rock is analyzed.The principal component regression,partial least squares regression and original variables are compared as input variables of the model It is found that thrust,rotational speed,torque and penetration rate are better as input variables of the model.On this basis,the real-time identification model of surrounding rock grade based on thrust,torque,rotational speed and penetration rate is given,and the identification accuracy is 85%.Then,the accuracy of the four machine learning models is compared with BPNN,BN,SVM and KNN.The accuracy of the four models is 75%,72.5%,77.5%and 52.5%respectively.The results show that the recognition accuracy of RF model is the highest.(2)Taking AR and SE as the evaluation indexes of TBM tunnelling performance,this paper establishes the surrounding rock classification system of TBM tunnel TP classification,and verifies the normal distribution of F,T,N,p,PR and U in all grades of surrounding rock,which shows that TP classification is reasonable.By analyzing the correlation between net driving speed and thrust under stability classification and TP classification,the results show that TP classification is obviously better than stability classification and is more suitable for guiding TBM construction.After that,the mapping model of F,T,N,p,PR,FPI and TP score of surrounding rock is established by BP neural network.The results show that the prediction accuracy is above 88%except for individual data.Finally,the SVM model based on MIV-BP is established,and the TP grade of surrounding rock is taken as the output variable.The results show that the prediction accuracy of SVM model is up to 95%,which can realize the real-time prediction of TP grade of surrounding rock.(3)A time series prediction model of TBM driving parameters based on LSTM network is proposed.By comparing the prediction results of LSTM model with the traditional regression models BPNN,GRNN and SVR,it is found that LSTM model is better in the prediction accuracy of all levels of surrounding rock and various excavation parameters.By comparing the prediction effect of LSTM model for different driving parameters under different surrounding rock grades,it is found that the higher the surrounding rock grade,the higher the prediction accuracy of the model;the prediction accuracy of construction speed is the lowest in all levels of surrounding rock,and the prediction effect of thrust and net driving speed is the best in all levels of surrounding rock.(4)Based on the RF model of real-time identification of surrounding rock grade and LSTM model of penetration rate prediction,an ant colony model is proposed to optimize the thrust and rotational speed of cutterhead with the maximum penetration rate as the optimization objective.Compared the optimization effect of exhaustive method model,simulated annealing model and ant colony model,the three models have almost the same optimization effect,but the calculation speed of ant colony model is the fastest,which is only 1/6-1/12 of exhaustive method;through the experimental verification of self-developed TBM simulation tunneling device,the results show that the thrust and speed optimized by ant colony model are the best combination of parameters,and the actual penetration rate is the best The predicted results are consistent with the ant colony model.