High-performance model predictive control methods for multilevel inverter-fed medium-voltage drive systems

Abstract

This dissertation presents comprehensive research on the modeling, control, and implementation of advanced multilevel inverter (MLI) topologies and model predictive control (MPC) strategies for medium-voltage (MV) drive systems. The primary objective is to achieve superior current tracking performance, reduced switching frequency, and minimized common-mode voltage (CMV), while maintaining low computational complexity. The research addresses critical limitations in existing MLI topologies and MPC methods such as high component count, increased cost and size, model inaccuracy, high computational burden through the development of novel converter configurations and control methodologies. A new five-level (5L) inverter topology is first proposed, featuring a reduced number of components and the elimination of multiple isolated DC-sources. The topology utilizes only flying capacitors (FCs) and switches, thereby reducing control complexity compared to existing 5L-MLI. A finite-control-set-MPC (FCS-MPC) method is also developed to control the proposed 5L-MLI, and the performance of the inverter is experimentally validated under various operating scenarios. Results demonstrate that the proposed inverter has superior harmonic performance and low switching power losses while operating at low switching frequency in comparison to the existing 5L-MLIs. Besides converter configurations, control methods play a pivotal role in system performance. Existing FCS-MPC are modeled based-on the forward Euler’s integration method due to its ease of implementation but suffer from significant prediction errors at larger sampling periods. To tackle this issue, a Heun integration-based-FCS-MPC approach is proposed for MLIs. The proposed method incorporates correction stage along with prediction stage to improve the prediction accuracy, resulting in a substantial reduction in current tracking error and switching activity. Experimental results confirm the effectiveness of the proposed approach through enhanced prediction accuracy while operating at a low switching frequency. To further tackle CMV and computational challenges, improved sequential MPC (SMPC) strategies are proposed. The proposed low-complexity SMPC eliminates the reliance on weighting factors and offline switching vector preselection to reduce the CMV. In addition, an enhanced sampled-data SMPC is proposed to improve the discrete-time model precision, significantly reducing current distortion and FC voltage ripple. Experimental validation on an MLI prototype demonstrates their excellent current regulation, lower CMV, and improved performance compared to existing SMPCs. Finally, an SMPC strategy with cost function-free current control and CMV mitigation is proposed based on the low-complexity SMPC framework. By directly determining the optimal voltage level from the reference AC currents, the proposed method removes the need for cost function optimization in the current control stage, while maintaining low CMV and reducing computational complexity. Experimental and simulation results demonstrate effective current regulation, low harmonic distortion, reduced FC voltage ripple, and satisfactory motor drive performance, confirming the practical suitability of the proposed SMPC for high-performance MLI-fed MV drive systems.