Optimizing Utilization of an MMSPC with Model Predictive Control

While the numerous degrees of freedom of a Modular Multilevel Series Parallel Converter (MMSPC) for automotive applications can be used for active balancing of the distributed batteries' energy and to reduce total converter losses, it is a challenge to determine the optimal next states.In this contribution, we present the implementation of a real-time model predictive control (MPC) algorithm for an automotive MMSPC that is able to optimize the switching states in each control cycle (12.5μs). The focus of the presented algorithm is the optimal control of the internal states of the MMSPC itself, without taking the control loops of the motor speed and torque into account. In addition to the capability of balancing all batteries within and between the phases simultaneously, the algorithm uses the common-mode (CM) voltage of the converter to reduce both the battery and the semiconductor losses. The control scheme is implemented on a field-programmable gPate array (FPGA) and validated on a machine test bench.