Computationally Efficient Direct Model Predictive Control with Increased Robustness for Medium-Voltage Power Conversion Systems

This thesis develops computationally efficient and robust model predictive control (MPC) methods for medium-voltage (MV) power electronic (PE) systems. As PE systems become more widespread across industries, limitations of conventional control approaches are emerging as a bottleneck in the development of reliable and sustainable power conversion systems. Specifically, existing control strategies often lack both high dynamic and excellent steady-state performance, especially in MV applications. A promising alternative is finite control set MPC (FCS-MPC), which enables PE systems to operate close to their physical limits while maintaining superior performance. However, the industrial adoption of FCS-MPC remains limited due to its high computational requirements and sensitivity to parameter variations.

The work addresses both challenges by developing a control algorithm that is applicable to both machine- and grid-side converters, with validation through real-time simulations and experimental tests. Case studies include an induction machine (IM) and a converter connected to an ideal or a distorted grid via an LC(L) filter.

It is shown that by reformulating the objective function of the optimal control problem underlying FCS-MPC, the number of candidate solutions can be reduced by several orders of magnitude. This results in a computational complexity reduction of over 99.9%, enabling the implementation of long-horizon FCS-MPC in real time without sacrificing performance. Compared to one-step FCS-MPC, the proposed approach can reduce the output current total harmonic distortion by 10% in an MV drive system with a three-level neutral-point-clamped converter and an IM, and by 30% when the same type of converter is connected to an ideal grid through an LC(L) filter. To ensure robustness, a computationally light parameter estimation algorithm is integrated into the control framework, providing estimates of key system parameters within ±5% of the true values with 95% confidence. Thus, the proposed FCS-MPC remains robust to unexpected variations in system parameters and retains favorable performance in both steady-state and transient conditions.