Abstract: Accurate prediction of mine water inflow is the core issue for water hazard prevention and control and protective water resource mining in coal mines. During high-intensity coal extraction, the dynamic evolution of permeability in the mining rock mass significantly impacts mine water inflow. Using the 5309 working face of a mine in Shanxi Province as the engineering case study, this paper comprehensively employs theoretical analysis, numerical simulation, and field measurements to reveal the spatial heterogeneity law of permeability in overlying strata disturbed by mining. An exploratory “Equivalent Large Well in Mining Rock Mass” method for the dynamic prediction of water inflow was proposed, which considers the dynamic changes in the equivalent permeability coefficient of the overlying strata. Subsequently, the dynamic prediction of water inflow was further implemented using the GMS platform, followed by engineering application and reliability verification. The results indicate: Coal seam mining induces a magnitude leap in the permeability of the overlying strata. The fracture aperture around the stope is two orders of magnitude higher than that in the central compacted zone. The permeability coefficient increased by up to 27 times post-mining. Furthermore, for an extended period after full extraction, the fracture network above the goaf maintains an “O-hape” spatial configuration characterized by high permeability at the periphery and low permeability in the central region. Based on the equivalent characterization of the permeability coefficient in the mining rock mass, the Equivalent Large Well in Mining Rock Mass method was proposed. This involves establishing a numerical “large well” model of water table drawdown that accounts for the dynamic variation of the equivalent permeability coefficient in the overburden. Using the propagation distance of water-conducting fractures breaking through key aquifers and the observed water inflow data from multiple mined-out working faces as the basis, and by dynamically associating the maximum/stable drawdown data at different stages, the working face water inflow is dynamically predicted, and the influence range of the drawdown “large well” is determined. Using this method, the predicted maximum working face water inflow was 30.71 m³/h, with a normal inflow maintained around 5.00 m³/h. During the prediction period, the water level at the center of the aquifer depression cone exhibited an evolutionary characteristic of “early-stage influence attenuation—rapid decline during mining—rapid rebound post-mining—slow recovery of recharge.” In contrast, predictions using initial hydrogeological parameters, which did not account for changes in overburden permeability, resulted in a prolonged water level fluctuation cycle in the overlying high-level aquifer, showing a “gentle decline—gentle rise” evolutionary pattern. Integrating the “three maps-two predictions” method and Transient Electromagnetic Method (TEM) geophysical prospecting, three water-rich anomalous zones were delineated above the working face. During directional drainage implemented in these anomalous zones, the dynamic recharge from the roof sandstone fissure water remained stable, with a maximum working face inflow of 29.0 m³/h and a normal inflow of 5.0 m³/h. Engineering application results demonstrate that the predictive model using the Equivalent Large Well in Mining Rock Mass method exhibits higher consistency with actual water inflow. These findings provide a theoretical foundation and engineering practice paradigm for roof water hazard prevention and control and the coordinated protection of water resources during mining in high-water-inflow mines in western mining areas.