Predictive control of modular multilevel series/parallel converter for battery systems

This paper presents a multi-step predictive control framework for modular multilevel converters with series/parallel connectivity - a more generalized topology of the H-bridge-based modular multilevel converters. The controller coordinates a wide range of objectives including converter output voltage, power losses, and balancing of the battery state of charge. Though capable of considering multiple steps ahead, the controller does not elicit a combinatorial explosion. The key idea is taking the converter's transistor currents, instead of switching signals, as the basic state variables. Based on these state variables, the converter can be modeled as a linear time-invariant system and the associated cost functions can be cast into linear-quadratic form. As such, the control problem is akin to the linear quadratic regulator, which allows efficient solution to the optimal control input with the prediction of multiple steps ahead. On an 8-module, 17-level battery system, prediction of three to five steps ahead reduces the power loss by 10% compared with predicting one step ahead, while further extending the prediction horizon does not bring improvement. The controller also shows good performance on balancing the state of charge of the batteries as well as distributing the losses among them.