Abstract: Being simple in principle and capable of controlling multiple objectives, the model predictive control (MPC) is more and more widely used in modular multilevel converter (MMC). However, the traditional MPC strategy has the problems of large computation and high switching frequency. An improved MPC method is thus proposed in this paper. Firstly, the current cost function is used to find the optimal combination of switch states; and then, the combination of switch states is optimized according to the input modular number on each sub-bridge arm at the adjacent time; finally, an integrated cost function is designed to control the capacitor voltage balance and eliminate the circulating current. The proposed method not only substantially reduces the calculation burden, but also reduces the average switching frequency of MMC and the power system's energy consumption, and is thus applicable to MMC system. An 11-level three-phase MMC-HVDC simulation system is established based on Matlab, and the simulation results have verified the effectiveness and feasibility of the proposed method.