Abstract: Aiming at the problem of roof collapse risk prediction of TBM construction coal mine inclined shaft crossing through weakly cemented strata, this study takes the TBM tunnelling passing through weakly cemented strata in Kekegai Coal Mine in Yulin, Shaanxi as the research background. The time series forecasting (TSF) algorithm of deep belief networks (DBN) is used to analyze the TBM construction parameter database and collapse events in weakly cemented strata inclined shafts. Evaluation indicators for TBM tunnelling roof collapse under such strata conditions are proposed for roof collapse risk prediction. The research results indicate that when using TBM method to construct inclined shafts crossing through weakly cemented soft rock strata in the western region of the Cretaceous, four indicators can be used for roof collapse warning, namely VAF<0.80, MRE>0.10, R₁<0.85, and R²<0.60; taking 12 sections from different positions of weakly cemented strata in the main inclined shaft to predict the risk of roof collapse, 10 locations are successfully predicted, achieving an accuracy rate of 83.3%, verifying the reliability of the collapse risk prediction algorithm.