A Method to Enable Reduced Sensor Capacitor Voltage Estimation in Modular Multilevel Converters

Abstract

ulk power applications such asshipping increasingly consider multilevel converter topologiessuch asmodular multilevel converters (MMC), which offers the advantages of scalability, good power quality, and reconfigurability. Theinternal functioningof MMCrequirescomplete knowledge of the capacitor voltages that make up their submodulesmeaninga large number of sensors areneededand thus a high number of potential points of failureexist. To increase reliability and reduce investment costs, state estimation techniques such as KALMAN filtershave been employedto replace the physical sensors. Analytical techniques based on the knowledge of arm current, arm voltage, and submodule states have also been developed. These techniques exploit the fact that at an insertion index of 1, the arm voltage equals the capacitor voltage on the submodulewhichpermitsthe estimation algorithm to refresh periodically with measured datatherebyincreasingtheaccuracy. This method requiresa long refresher time, especially when many submodules are used per arm. In this study, we propose an improved analytical estimationby not only usingunity insertion indices,but also exploiting transitions between two successive insertion indices.The study was carried out on a 4 submodule per arm MMC system. The estimated capacitor voltages were then compared with sensor-basedvoltage measurementsconfirming the validity of the proposed method. It was then integrated into a complete MMC controller including the inner controlssuch ascirculating current and capacitor voltage balancing