Application of a model predictive control method with reduced computational load for a three-level T-type inverter

Voltage source inverters have been extensively researched in a wide range of areas from electric vehicles to motor drives, from uninterruptible power supplies to grid integration of renewable and sustainable energy sources. Inverters, which act as a bridge between energy sources and loads, are expected to meet requirements such as efficiency, harmonics, electromagnetic interference, controllability, and reliability. In low voltage applications, the three-level T-type inverter structure combines the advantages of the widely used two-level inverters such as simple operating principle, and low conduction losses with the benefits of three-level inverters such as low harmonic content generation in output voltage and current, reduced dv/dt ratio, reduced voltage stress on switches, low common mode voltage, reduced switching losses. In addition to the structure of the inverter, the control method to be applied to the inverter also plays a key role in performance criteria. Classical control methods are usually tuned to achieve satisfactory performance over a narrow operating range. However, the model predictive control method, which has become popular in recent years, can control complex and nonlinear system dynamics, while making it possible to effectively add design criteria to the control structure. With microcontroller technology, which has increased computational capability, the model predictive control method has gained a prominent place in power electronics applications. However, the computational burden of the model predictive control method increases due to the increasing number of switchings in multilevel inverters. In the literature, reducing the computational load of the model predictive control method while preserving its superior features has attracted the attention and interest of researchers. In this thesis, an algorithm that divides the space vector diagram into specific regions and uses a limited number of candidate vectors is proposed to reduce the computational burden of the classical model predictive control method. The proposed algorithm includes two strategies, each using different candidate vector groups, which evaluate criteria such as balancing the neutral-point voltage and limiting the common-mode voltage. This model predictive control method is applied to a three-phase, three-level T-type inverter with SiC MOSFET semiconductor switches. An experimental prototype is built to verify the effectiveness of the proposed algorithms. Their behaviors in steady-state and dynamic conditions are investigated, and their performances are analyzed according to relevant international standards. Of the proposed algorithms, the 8V-MPC method reduces the computational load by 54.2% compared to the classical model predictive control method, while another approach, the 7V-MPC method, achieves an even higher reduction of 58.8%. Finally, both approaches show similar or superior performance in criteria such as total harmonic distortion, common-mode voltage, and neutral-point imbalance compared to the classical method. With the proposed approaches with reduced computational load, it becomes possible to add new constraints or objectives to the model predictive control method.