Accurate and fast simulation of junction temperature is crucial when using optimization algorithms to improve the heat performance of power modules. Due to its coding characteristics that are easy to integrate into optimization algorithms, this paper uses an improved lattice Boltzmann method to simulate the thermal distribution of a power module with a pin-fin heat sink in its working state. Due to the large computational resources and slow speed required for 3D models, this paper also proposes a method of combining different models. The simulation results have good convergence and can obtain the highest junction temperature, average junction temperature, and inlet and outlet water temperatures. The feasibility of this simulation method was verified by comparing the results with CFD simulation software.