In order to realize the accurate and quick evaluation on the surrounding rock mass in front of tunnel face during the construction process, in this paper, a dynamic classification method for tunnel surrounding rock, which is based on traditonal BQ classification method, is proposed depending on machine learning and reliability algorithm. The machine learning tool is selected as the least squares support vector machine(LSSVM), and its parameters are optimized by the bacterial foraging algorithm(BFOA) to construct a nonlinear mapping relationship between the hierarchical index group and the surrounding rock level. The grading index group is made up of parameters include geological advance prediction results and the strength rebound value of the face surface, which are easy to be acquired during the construction process. Furthermore, the reliability theory is applied to verify the randomness problems that may exist in the results of some grading indicators. By constructing the functional function of reliability calculation through machine learning results, the surrounding rock classification with probability meaning are realized. To verify its feasibility in some sections based on calculation results, the new dynamic grading method is applied in Zhenfengling tunnel and its applicability is proved by automated monitoring data. The results show that the classification method can effectively realize the dynamic grading calculation of surrounding rock during construction, which provides a new idea for the dynamic design process of tunnel construction.
