Aiming at the problems that the wind cover of the amplitude-flow fan in rail transit restricts its flow and efficiency, and the optimal design of the amplitude-flow fan depends on high-precision simulation and long design period. In order to shorten the R&D and design cycle of the cross-flow fan and improve the ride comfort of subway cars, a structural optimization method based on the Kriging surrogate model was proposed. Design points were selected via Latin hypercube sampling, and a Kriging surrogate model was constructed using numerical simulation results. Multi-objective optimization was performed on the surrogate model using the NSGA-II algorithm. The entropy weight TOPSIS method was then applied to rank the relative closeness of 100 candidate solutions, from which the optimal design was obtained. The optimized left and right housing radii were determined as 128.37 mm and 58.67 mm, respectively, with an outlet width of 36.45 mm and an inlet angle of 198.57°. Numerical simulation results show that the optimized fan exhibits reduced tip leakage flow, suppressed flow separation, and more uniform static pressure distribution. The flow rate, total pressure efficiency, and maximum flow velocity of the fan are increased by 16.34%, 4.36%, and 18.32%, respectively. Experimental test results indicate that, due to idealized simulation conditions, the measured flow rate and total pressure efficiency are slightly lower than the simulated values. Future research will explore ways to improve the prediction accuracy of the surrogate model by introducing global correction factors and establishing error models.