Research On TBM-Rock Interaction Model And Optimal Decision Method Based On Multi-Algorithm Fusion

The TBM construction method has become the primary choice for the construction of deep and long tunnels at home and abroad with the advantages of fast construction speed,high hole quality,and eco-friendly environment.However,TBM is sensitive to geological conditions,and the interaction mechanism between rock mass and cutterhead is complicated,which makes the perception of geological information and the selection of tunneling parameters during the construction process are particularly important for safe and efficient tunneling of TBM.In response to the urgent needs of TBM intelligent tunneling,this paper focuses on the key scientific problem "Rock-TBM interaction model representation and excavation parameter intelligent decision-making method",and artificial intelligence methods are used to mine TBM field data relying on the Jilin Songsong Water Supply Project.The machine-learning based rock-TBM interaction models are established,and the optimization and decision-making methods of TBM tunneling control parameters are proposed.The main research contents and results are as follows:(1)The TBM field data is divided and extracted according to the ascending segment and the steady-state segment,and the abnormal values in the original data are filtered and corrected based on the absolute median and Mahalanobis distance.The on-site tunnel geological survey data was matched with the TBM excavation data,and a rock-TBM interaction database with a sample size of 4459 was established,including field data such as the grade of surrounding rock and the main TBM excavation parameters.The database provides a data basis for the establishment of rock-TBM interaction models and optimization decision-making methods for tunneling control parameters.(2)A rock mass class prediction model based on Ada Boost-CART is established,which can predict the rock mass class of the tunnel excavation face in real time according to the characteristics of the TBM parameters.The application of SMOTE reduces the impact of the imbalance of surrounding rock mass classes on the prediction performance of the model.The prediction accuracy of the model is verified by actual engineering data.Compared with the ANN,SVC and KNN models,the comprehensive performance of Ada Boost-CART is better.The analysis of the importance of the model features shows that the average cutter head speed has the greatest impact on the prediction results of surrounding rock grades.(3)The predictive model of TBM tunnelling control parameters based on the QPSOANN algorithm is established,which realizes the prediction of the steady-state cutterhead speed and the mean value of penetration.Compared with BPNN,QPSO-ANN effectively improves the global optimization capability of the model.The tunneling control parameter optimization model based on ILF-QPSO-ANN is established with improved loss function of QPSO-ANN.The test results showed that the output result of ILF-QPSO-ANN can effectively reduce the specific energy of cutterhead and increase the advance rate of TBM.(4)The selection and adjustment of tunneling control parameters is transformed into a multi-objective optimization problem.Taking empirical error,rock breaking specific energy and net excavation rate as sub-objectives,the DE-NSGAII and GPR proxy models are used to obtain the non-dominated set composed of ILF-QPSO-ANN with different weight coefficients.Considering factors such as TBM construction,geological conditions and geological problems,the analytic hierarchy process and anti-entropy method are used to calculate the weight of each sub-objective.The optimal combination of weighting and TOPSIS are used to make decisions on non-dominantly concentrated excavation plans,and a multi-objective and multi-modal excavation control parameter optimization decision method under different rock mass classes is proposed.