Abstract: The three-level converter has become the mainstream converter topology because of its good output power quality and high power factor. The three-level dual pulse width modulation frequency speed control system with rectifier–inverter structure has become a research hotspot in academic circles because of its advantages of bidirectional energy flow, high power quality, and controllable intermediate direct current (DC) voltage. Aiming at the high-performance and high-efficiency control of the three-level neutral point clamped (NPC) rectifier–inverter drive system of an induction motor, this study builds a prediction and loss model of the three-level rectifier–inverter system, constructs a cost function including the midpoint voltage balance and loss optimization, and proposes a model predictive control without weighting factors based on a sequential parallel structure. With the development of the field of power electronics, the control performance and efficiency of the converter have gradually improved, and the model predictive control applied to the converter is no longer limited to the traditional control objectives. The proposed strategy introduces the DC bus midpoint voltage and converter switching frequency control to the traditional sequential model predictive control and constructs a unified cost function with multiple control objectives. According to the actual requirements of the multiple control objectives in the operation of the rectifier–inverter system, the control objectives in the cost function are divided into primary and secondary control objectives and classified into two sequence optimization sets, and different sequence sets are sequentially optimized. In the same sequence set, adaptive parallel optimization is used to select the optimal switching state, which ensures the synchronous optimization of each control object in the sequence, thus avoiding the need for weighting factors. The parallel structure can rank multiple control targets, reduce the number of sequences to increase the number of optional voltage vectors between each sequence and increase the control effect of nonprimary control targets. Moreover, synchronous optimization of control targets of the same importance is realized at the same level, which solves the problem that targets of similar importance must be sequentially optimized in conventional sequential model predictive control, solves the problem that the priority of different targets is difficult to adjust, and has stronger applicability for the complex topology structure with multiple control requirements. The simulation and experimental results showed that the proposed algorithm can improve the steady-state and dynamic performance of the system, reduce the midpoint voltage bias, reduce the switching frequency of the rectifier and inverter, reduce the total harmonic bias, and adjust the midpoint voltage unbalance.