To enhance the accuracy of motor driver thermal simulations under challenges of numerous hard-to-measure parameters, this study proposes a parameter correction method integrating experiments with FLOTHERM simulations. By analyzing steady-state temperature data from driver boards, it identifies key parameters through Latin hypercube sampling and Pearson correlation-based sensitivity analysis. Strongly correlated parameters are then optimized using response surface modeling and genetic algorithms. Results confirm that this data-driven approach achieves accurate thermal modeling with limited simulations by effectively combining experimental measurements and computational optimization.