The introduction of cavitation effect in ultrasonic rolling is expected to further improve the reinforcement quality. In order to maximize the cavitation effect, the optimized design of the flow field structure in ultrasonic rolling was carried out. Firstly, software Fluent was used to simulate the cavitation in ultrasonic rolling, and the vapor volume fraction (VVF) at three key locations around the roller was determined. Secondly, the optimal Latin hypercube method was used for the experimental design. The flow field structure parameters were used as the optimization variables, and the VVF at three key locations around the roller was used as the optimization target. The VVF approximate models were established with the second-order response surface method. Then, the weight values of each VVF were determined by combined AHP and entropy weight method, and the optimal flow field structure parameters were obtained by optimal solution of the approximate models using genetic algorithm NSGA-II. Finally, comparison of the VVF obtained with the optimized structure and the initial structure was performed to verify the optimization results. The results showed that the second-order response surface approximate models constructed for VVF fit well based on the results of 20 optimal Latin hypercube tests, and VVF at all three locations passed the significance test at 95% confidence level. After comprehensive analysis, the weights of the three optimization objectives were determined as 0.2791, 0.2516 and 0.4692. Compared with the initial structure, the VVF at the three key positions of the optimized structure was increased remarkably by 21.6%, 156.4% and 44.1%, respectively. The optimized flow structure can be applied to the ultrasonic rolling system to improve the cavitation effect.