Abstract: In multi-physics coupled simulations of nuclear reactors, the accuracy of the Monte Carlo program significantly influences the confidence of coupled calculations. To enhance the precision of Monte Carlo simulations while maintaining compatibility with the coupling process, this paper develops a dynamic global weight window variance reduction method within the Monte Carlo program RMC. This method dynamically updates and applies global weight window parameters based on the neutron flux distribution and statistical error distribution for each active generation in Monte Carlo critical calculations, without modifying the original coupling workflow. The effectiveness of the variance reduction method is verified through critical calculations of the Hoogenboom-Martin full-core benchmark and the plate-fuel assembly core benchmark. The results demonstrate that the dynamic weight window method effectively reduces the global statistical error in the core. Additionally, RMC-COMSOL nuclear-thermal coupling simulations using plate-fuel-assemblies are conducted for validation. After applying the dynamic weight window, the power distribution differences between iteration steps are significantly reduced, which effectively improved the convergence of the neutronics/thermal-hydraulic coupling simulations.