The power electronic transformer （PET） based on modular multilevel converter （MMC） is prone to failures and disturbances under non-ideal operating conditions, which seriously affects the system power quality. Aiming at the poor dynamic and steady state performance of traditional control methods under non-ideal operating conditions, a continuous control set model predictive control strategy based on self-adaptive auto-disturbance rejection proportional integral controller is proposed in the MMC-PET rectifier stage. Firstly, a self-adaptive auto-disturbance rejection proportional integral controller is designed to solve the problems of poor signal tracking and disturbance rejection in the voltage outer loop. Secondly, the continuous control set model predictive control method is used in the current inner loop to improve the response speed and steady state performance of the system, and an improved carrier phase-shift modulation strategy is introduced to solve the problem of bridge arm current distortion. Finally, the MMC-PET system is compared by simulation and experiment under non-ideal operating conditions, such as sudden change of loads on the grid side, unbalanced grid voltage, and load input at the output stage, to verify the superiority of the proposed control strategy.