Deep belief network for roof instability forecasting in TBM-excavated inclined shafts of weakly cemented strata

The collapse of roof strata in coal mine roadways significantly contributes to the instability of surrounding rock. Existing technologies for predicting the risk of roof collapse in inclined shafts constructed with Tunnel Boring Machines (TBM) mainly rely on statistical regression analysis or shallow machine learning methods. However, these approaches often inadequately capture nonlinear features and neglect the importance of temporal correlations. In this paper, we present a deep learning algorithm tailored for predicting roof collapse in coal mine roadways, focusing on the weakly cemented strata encountered during TBM excavation at the KeKeGai Coal Mine in YuLin, Shaanxi Province. We employ the Time Series Forecasting (TSF) algorithm based on Deep Belief Networks (DBN) to analyze a database of TBM construction parameters and past collapse events associated with these weakly cemented strata. Our study proposes specific evaluation indicators to assess the risk of roof collapse under such geological conditions. The results of our research indicate that when constructing inclined shafts through weakly cemented soft rock strata of the Cretaceous period in the western regions using the TBM method, four key indicators can effectively serve as early warning signs for potential roof collapse: VAF < 0.80, MRE > 0.10, R 1 < 0.85, and R 2 < 0.60. We also predicted the risk of roof collapse for twelve segments of strata at various locations in the main inclined shaft, successfully forecasting ten instances and achieving an accuracy rate of 83.3%. This validates the reliability of our proposed algorithm for predicting the risk of roof collapse.