The invention discloses an NPC three-level rectifier optimization finite control set model prediction method, and relates to the field of power electronics and power transmission. According to the method, firstly, a three-level PWM rectifier discrete prediction model is established and initialized, state sampling is carried out on a PWM rectification system, then a predicted value and an expected value of a system state are calculated according to a recorded optimal switching vector, finally, whether a current error and a neutral-point potential error meet reference deviation at the same time or not is judged, and finally, whether the current error and the neutral-point potential error meet reference deviation or not is judged. If yes, recording the switching vector at the previous moment, and outputting the switching state; if not, the reference voltage is calculated, the sector where the reference voltage is located is judged, the value function of the switching state of the sector is solved, the switching vector with the minimum value function is recorded, and the switching state of the switching vector is output. The method has the advantages that when the current error and the neutral-point potential error are controlled within a certain range, the average switching frequency of the converter can be reduced, when the current error and the neutral-point potential error exceed the range, 27 times of traversal are reduced to 10 times, the calculation burden is reduced, and the method is suitable for being applied to a high-power PWM rectifier.