In a fusion reactor, an in-vessel loss of coolant accident (LOCA) involving the helium-cooled blanket will lead to the resuspension and migration of tritiated and micron-sized tungsten dust deposited at the bottom of the vacuum vessel under high-speed helium jets, significantly increasing the radiological release hazard. Drag force dominates particle transport dynamics, yet conventional computational fluid dynamics (CFD) fails to accurately characterize particle-fluid interactions in high-speed and rarefied flow fields. This study proposes a hybrid framework combining CFD with the Maxwell slip boundary and direct simulation Monte Carlo to simulate micron-particle drag force during in-vessel LOCA. The Cunningham drag correction formula accounting for wide-range Knudsen number (Kn) rarefied helium flow (0.01 < Kn < 2 × 104) was established using nonlinear least-squares fitting. Furthermore, by incorporating compressibility effects induced by high-Mach flow, a symbolic regression-based drag model was established as a function of Mach number (Ma) (0.033 < Ma < 1.31) and particle Reynolds number (1 × 10−4 < Rep < 50). This model provides critical engineering guidance for predicting radioactive dust migration in fusion reactors and the proposed framework could be extendable to loss of vacuum accidents and other engineering applications involving particle drag in high-Mach rarefied gas flows.