Abstract: This study investigates the application of laser welding technology in the precision manufacturing and mass production of metallic bipolar plates for fuel cells, where its high accuracy and efficiency offer significant advantages. However, challenges persist in welding ultra-thin metallic bipolar plates, including inconsistent weld quality and thermal deformation issues. To address these challenges and optimize the welding process window more effectively, we developed a compound Gaussian surface-cylinder heat source model tailored for partial-penetration welding of ultra-thin metallic bipolar plates using COMSOL Multiphysics. A full factorial design involving four parameters at five levels was established based on penetration depth, weld width, and joint width measurements. Subsequently, a BP neural network was trained using actual weld pool morphology data through JMP software to calibrate the simulation model. This calibration process significantly enhanced the correlation between simulated weld pool profiles and experimental observations. Comparative analysis demonstrated that simulations using the refined laser heat source model achieved relative deviations within ±5% when validated against empirical welding experiments on bipolar plates. These results confirm that the proposed model provides a reliable theoretical framework for optimizing laser welding processes in metallic bipolar plate fabrication.