Abstract: An intelligent surrounding rock grade identification method based on the fusion of drilling parameters and tunnel seismic prediction (TSP) method is proposed to improve the identification accuracy of surrounding rock grade based on geological information in tunnel survey and construction stages. Drilling parameters, TSP reports, and geological sketches are collected from the construction site to interpret corresponding data and surrounding rock information. Then, five fusion scenarios are set and four machine leaning classification algorithms, i.e., K-nearest neighbor, gradient boosting tree, random forest, and extreme tree, are employed to verify the separability of fused data for determining the final fusion mode of the drilling parameters and TSP data. To determine the impact of various data partitioning methods on the fitting ability of models, 500 partitioning modes are applied to the dataset by setting random number seeds to fully fit each model. Results show that the extreme tree model yields the best surrounding rock grade classification effect with a maximum accuracy of 91.3% by fusing drilling parameters such as feed rate, thrust pressure, impact pressure, and rotation pressure with the longitudinal and transverse wave velocities in TSP, respectively. Finally, the Bayesian optimization algorithm is used to optimize the hyperparameters of the extreme tree model. The optimized model exhibits improved classification performance for various rock grades, with overall and increase accuracies of 93.27% and 1.93%, respectively.

Key words: tunnel, rock mass classification, drilling parameters, tunnel seismic prediction, machine learning, Bayesian optimization