Research on optimization strategy of TBM tunneling parameters based on stratum perception and simulation tunneling experiment

• Published in: Tunnelling and underground space technology 2024-05, Vol.147, p.105743, Article 105743
• Main Authors: Fu, Kang, Qiu, Daohong, Xue, Yiguo, Tao, Yufan, Kong, Fanmeng
• Format: Article
• Language: English
• Subjects: Machine learning; Optimization strategy; Simulation tunneling experiment; Stratum perception; TBM
• ISSN: 0886-7798
• DOI: 10.1016/j.tust.2024.105743

•The quaternion classification problem of surrounding rock grade was converted into the binary classification problem.•Based on the data balance method, the EasyEnsemble model of TBM tunnel surrounding rock stratum perception was established.•A flow frame for optimization of all grades surrounding rock tunneling parameters was proposed.•The difference of optimization results among exhaustive method, simulated annealing algorithm and ant colony optimization algorithm was compared.•The simulation tunneling experiment was carried out using the self-developed model test equipment, and the optimization results of the ant colony optimization model were verified. Real-time perception of surrounding rock stratum and optimization decision of tunneling parameters are of great significance for TBM safe and efficient tunneling. Taking a diversion tunnel in Xinjiang as the engineering background, 948 groups of tunneling data of surrounding rocks at all grades were collected. Aiming at the problem of unbalanced tunneling data of different grades of surrounding rock, the quaternion classification problem of surrounding rock grade is converted into the binary classification problem, and the surrounding rock stratum perception model is established based on EasyEnsemble algorithm. On this basis, a flow frame for optimization of all grades surrounding rock tunneling parameters is proposed, and the differences between the optimization results of exhaustive method, simulated annealing algorithm and ant colony optimization algorithm are compared. The results show that the prediction error of ant colony optimization model is less than 5 %, and the running time is only 1/12 ∼ 1/6 of the exhaustive method. Then, based on the self-developed model test equipment, a similar materials test block of surrounding rock stratum is configured, and a scale cutter head is used for simulation tunneling experiment. The experiment shows that the ant colony optimization model optimization result is the best combination of tunneling parameters, which shows that the proposed optimization flow frame of tunneling parameters has good applicability. Finally, the AdaBoost model was trained with unbalanced samples of various grades of surrounding rock, and the prediction results were statistically analyzed. The results showed that the AUC value of EasyEnsemble model increased by 11.33 % on average, the F-measure increased by 18.12 % on average, and the G-measure increased by 14.00 % on average compared w