Finite control set model predictive control (FCS-MPC) is widely used in modular multilevel converters (MMC) because of its ability to achieve multi-objective control. As the number of submodules increases, model predictive control (MPC) faces exponential computational growth, high computational complexity, and difficulty in integrating the weighting factors. To solve the above problems, a multi-variate adjusting set model predictive control (MAS-MPC) method is proposed. This method performs a fast correction of the submodule control set by output current and arm voltage difference, and evaluates two cost functions to obtain the optimal switching vectors. In addition, a capacitor voltage balance algorithm based on differentiated medium term is proposed. This can effectively reduce the complexity of the sorting algorithm. To verify effectiveness, a three-phase MMC system with 10 voltage levels is built based on Matlab/Simulink, and compared with the traditional strategy. The proposed strategy drastically reduces the computational effort of the system while reducing the harmonic content of the output current and the circulating current. A system with faster dynamic response is achieved.