Abstract: Jamming and shutdown of tunnel boring machines(TBMs) and rock collapse often occur during tunneling due to unknown geological conditions. Therefore, the main mechanical parameters, such as rock mass integrity index, uniaxial compressive strength, internal friction angle, cohesion, deformation modulus, Poisson′s ratio, robustness coefficient, and elastic resistance coefficient, are considered as references. The fully adaptive noise ensemble empirical mode decomposition (CEEMDAN) and time convolutional neural network (TCN)are used to establish a method for predicting the grade of the surrounding rock according to the principle of "decomposition first and then reconstruction". The results reveal the following: (1) The mean square error between the predicted and actual values of the surrounding rock grade based on the CEEMDANTCN combined model is <0.07, the root mean square error is <1.67, the average absolute percentage error is <0.45, and the average absolute error is <0.14, with a fitting coefficient of 95.2%. (2) The CEEMDANTCN combination model has the advantages of low error, good fitting effect, and high practicality, accurately predicting the grade of surrounding rock at the tunnel face and achieving intelligent classification of surrounding rock types, which is crucial for efficient TBM tunneling and early risk warning.

Key words: tunnel surrounding rock, prediction and evaluation, CEEMDANTCN, quality grade of surrounding rock, deep learning