In a single-phase three-level inverter system, the classic finite control set model predictive control cannot guarantee the output power quality at a lower sampling frequency. To solve this problem, a multi-state virtual-vector model predictive control method is proposed. In a single control period, multiple switch states are output to improve the power quality of the grid current. The inverter predictive model is established, and a set of switch states are initially selected according to the predicted vector, and the work time of each state is determined by the designed current cost function. To effectively balance the voltage of the DC bus, a voltage balancing cost function is designed, and the switch state is further adjusted to obtain the final switch state sequence output to the inverter. A simulation is built and the analysis is carried out. The results verify the effectiveness of the proposed control strategy, the grid-connected power quality is effectively improved, and the voltage fluctuation of the DC bus is small. This work is supported by the National Natural Science Foundation of China (No. U1504617 and No. 61803145) and the Project of Henan Science and Technology Department (No. 192102210231).