Abstract: The rapid development of China’s nuclear power industry, encompassing both domestic innovation and international exports, has created an urgent need for localized nuclear power software solutions. As essential components of such software, system analysis codes have seen significant development efforts by Chinese research institutions, primarily based on two-fluid models similar to those used in RELAP5 and TRACE. However, these codes require continuous validation and improvement to meet evolving industry demands. Internationally, the adoption of three-field models that explicitly account for droplet behavior (as exemplified by codes like SPACE and CATHARE3) has become a prevailing trend to enhance simulation accuracy, particularly for complex thermal-hydraulic phenomena. Building on this global momentum, the NUSOL (Nuclear Safety and Operation Research Laboratory) team at Xi’an Jiaotong University previously developed a three-field model-based system analysis code. This study presents systematic optimizations to the code’s critical components, with particular emphasis on reflooding scenarios where droplet dynamics significantly influence thermal-hydraulic behavior. The optimization process began with sensitivity analysis-guided improvements to phase transition criteria and heat transfer modeling. A key focus involved refining the transition criteria between film boiling and transition boiling heat transfer modes, crucial for accurate prediction of heat transfer coefficients and wall temperatures. Through analysis of UC-B reflooding experimental data encompassing four distinct conditions of reflooding rates (2.44-12.67 cm/s) and temperature (23.3-67.8 ℃), the multiple minimum film boiling temperature models were evaluated. Comparative analysis revealed that the Berenson model demonstrated superior accuracy in three-field simulations, and was subsequently integrated into the code’s framework. Further optimization addressed the computational errors observed under low reflooding rate conditions, where existing three-field implementations showed particular deviations. By revising the film boiling heat transfer models for both liquid and vapor phases at the wall interface, the code’s capability to capture subtle phase interaction mechanisms is enhanced. The improvements were validated against FLECHT SEASET rod bundle reflooding experimental data, enhancing the model’s accuracy in low reflooding rate conditions for key parameters including quench front progression and peak cladding temperatures. The wall heat transfer and film boiling determination criteria in the three-field model are successfully optimized, the improvements are verified through reflooding experiments. The results demonstrate that the improved code accurately simulates the thermal-hydraulic behavior of droplets during reflooding, significantly improving the precision of system analysis codes. The improved code provides a more reliable tool for safety analysis of advanced reactor design. These advancements play an advancing role in accelerating the localization and innovation of China’s nuclear power system code.