Voltage Estimation for Diode-Clamped MMCs Based on a Simplified Neural Network
The modular multilevel converter (MMC) is a popular solution in high-voltage dc application and has significant potential in others. Generally, the MMC's stable operation is at the expense of numerous sensors, communication burdens, and complicated balancing strategies that can suppress its expansion in to cost driven applications. Hence, the introduction of a sensorless voltage balancing strategy with a simple controller is an attractive objective. A diode-clamped MMC offers a simple and yet effective solution by providing a balancing path between two modules through a diode. However, to compensate the lack of bidirectional energy transfer, modifying modulation technique is necessary. The level-adjusted phase-shifted carrier (LA-PSC) modulation introduces a small circulating current that ensures a correct balancing direction. Although the open-loop implementation of LA-PSC might be necessary for cost reduction in some applications, protection and control considerations may still necessitate careful monitoring of the modules' voltages. This paper proposes a voltage estimation strategy based on a simple neural network that does not require any measurement of the modules' voltages. Provided analysis as well as the simulation results confirm that the estimator can track the voltages with above 99% accuracy during balanced and imbalanced conditions.
