Abstract: To investigate the dynamic seepage mechanisms of fluids within fractures of tight reservoirs, a three-dimensional digital core fracture structure of an actual reservoir was constructed based on deep learning segmentation results. First, the fracture connectivity was evaluated. Then, single-phase flow permeability simulation was conducted, and gas-water two-phase flow displacement was studied using a level-set method coupled with Navier-Stokes (N-S) equations, with solutions obtained using the finite element method. The results showed that the deep learning method efficiently and automatically segmented fractures in core images with an accuracy of 85%. Connected fractures played an important role in rock permeability. Different fluid properties affected flow pressure and velocity, thereby affecting permeability. During the displacement simulation, the distribution characteristics of gas and water phases were clearly observed. As the displacement progressed until seepage completion, the fluid saturation in narrow fracture channels remained nearly unchanged, serving as the primary storage space for residual gas phase. Fractures with relatively good connectivity, characterized by great width and straightness, became the main seepage channels where gas recovery rates tended to stabilize. The research findings provide guidance for studying gas-water two-phase flow in fracture spaces of tight reservoirs under microscopic conditions.