Flywheel energy storage system (FESS) high-speed permanent magnet synchronous motors (HPMSM) often use magnetic bearings. The motor rotor is susceptible to current ripples, resulting in vibration noise and even instability. In this paper, an improved model predictive control strategy is proposed to reduce the ripple current of the bus capacitor while optimizing the current of the HPMSM to reduce the vibration noise. Firstly, a three-vector finite set model predictive control algorithm is designed to solve the problem of large current ripple caused by the limited number of optional vectors in the control period of the traditional finite set model predictive algorithm. In order to reduce the number of switch state transitions and reduce the switching loss, a zero-vector optimization allocation strategy is proposed, which the zero vector at the end of the current control cycle is the same as the zero vector at the beginning of the next cycle. Then, according to the switching state of the motor side and the grid side converter, the bus capacitor current is predicted, a unified cost function including the bus capacitor current is established, and the optimal vectors of the motor side and the grid side are selected by comprehensive evaluation. Finally, the correctness is verified by simulation.