Abstract: To improve the accuracy of the intelligent classification model based on drilling parameters, the interaction between drilling parameters and the geological heterogeneity of the surrounding rock is taken into account. Key variables such as penetration velocity, feed pressure, hammer pressure, and rotation pressure are used as the original drilling parameter features. These features are then integrated into a multivariable drilling parameter characteristic system using feature combination and statistical methods. Six machine learning methods are applied for rock classification, with a comparative analysis conducted on the interclass distances of rock grade samples and the accuracy of the classification models, both before and after feature extraction. The results reveal significant improvements with the multivariable feature system. The average interclass distance of different rock grade samples increases by 66. 09% to 85.41%, while the overall accuracy of the classification models rises from 75.5%－87.5% to 90.0%－92.5%. These findings demonstrate that the multivariable drilling parameter feature system can significantly improve the performance of rock classification.

Key words: tunnel surrounding rock classification, drilling parameters, feature mining, drilling energy index, machine learning