The construction of railway tunnel in carbonaceous slate environment is easy to cause rock mass disturbance, which leads to large deformation of surrounding rock and then threatens the safety of tunnel construction and operation. Therefore, based on the Wanlamu tunnel and Huajiaopo tunnel projects of Lixiang railway, combined with the regional geological conditions of the tunnel, and using the field monitoring technology and statistical principle, this paper analyzes the characteristics of tunnel surrounding rock pressure, secondary lining cracking, and initial support deformation under the conditions of different bedding slate. It is analyzed that the large deformation of carbonaceous slate surrounding rock is mainly related to the bedding, characteristics of surrounding rock, bias pressure, and the existence of high ground stress. The tunnel deformation is mainly manifested in peripheral convergence and vault settlement, and the peripheral convergence value is greater than the vault settlement value. In this study, the maximum daily convergence deformation of the tunnel peripheral convergence exceeds 117.8 mm, and the maximum convergence deformation can reach 951.7 mm, which is seriously affected by the slate bedding. According to statistics, more than 90% of the cross-section surrounding rock deformation of the tunnel with the slate bedding angle of 30° exceeds 600 mm. Then, based on BP neural network, the large deformation prediction model of railway tunnel under the carbon slate environment is constructed, and the accurate prediction of large deformation of carbon slate tunnel is realized, with the minimum prediction error of only 0.81%. It is pointed out that the main factors leading to the large deformation of the carbon slate tunnel are the lithology of the slate, the difficulty in predicting the deformation of the surrounding rock, and the failure of the initial support to close in time. The tunnel deformation can be effectively controlled by strengthening the geological work, optimizing the design parameters (enhancing the support stiffness, multiple support, and increasing the reserved deformation), and mining active reinforcement.