Tunnels in karst areas are prone to collapse during construction. There are many analyses on the mechanism of tunnel collapse in mechanical aspects, but the mechanism of tunnel collapse in karst weak fracture zones and other strata were seldom examined. In order to ensure the safety, economy and feasibility of tunnel construction, it is necessary to master the mechanism of collapse in the tunnel construction. Relying on a tunnel project in a karst broken stratum in Guizhou, where the collapse phenomenon occurred during the excavation process, the monitoring data of the tunnel are examined, and the construction principle of the BP neural network is used to invert the stratum parameters of the tunnel. The inversion soil mechanical parameters are input into different construction models constructed by using the FLAC3D finite element software, and the collapse failure mechanism and risk of typical sections are judged and analyzed. The results show that the construction method has a great influence on the stability of the tunnel excavation, and for the tunnel with the surrounding rock grade V , the three-step seven-step method and the single-side wall pilot pit method are safer for construction, and the tunnel collapse has no relationship with the simultaneous excavation of the tunnel in both directions. The predicted value of the tunnel vault displacement obtained by the inversion is 2.3 cm, and the predicted value of the surface displacement is 1.2 cm. The deviation from the monitoring data is about 13%, and the inversion result has certain reliability. The research results are of important guiding significance for the construction of tunnels and highways in weak and broken strata in karst areas.