Abstract: Existing model predictive control methods for modular multilevel converters need to establish cost functions and have a complex calculation process. This paper proposed a backward prediction based model predictive control strategy for the MMC, which utilized forward difference method and nearest level approximation to predict the optimal switching state directly, in order to control the phase current and circulating current. Also hysteresis control was introduced in circulating current control to improve the output power quality and reduce the switching loss. An MMC simulation system with 20 submodules in each arm was built to verify the proposed theory. The simulation results show that the proposed strategy is calculating-convenient and has good performance in output characteristics and dynamic response characteristics with significantly lower switching loss.