Real-Time Implementation of Predictive Control in Power Inverters Based on Nearest Neighbor Searching

Model predictive control is a powerful methodology to control multilevel power inverters because of its ability to deal with multiple variables and control objectives but presents some difficulties, such as high computational burden and the fact that the control input is held constant during each sampling period. Multirate model predictive control was later proposed to alleviate the latter issue but would become even more complex to compute. This work proposes the implementation of multirate model predictive control using a novel adaptation of nearest neighbor searching to learn the control law. This effort made it possible for this control technique to be implemented in real time by taking advantage of hardware-accelerated parallelization. The method is tested on a five-level diode-clamped converter operating in inverter mode. The algorithm was implemented in a field-programmable-gate-array-in-the-loop experiment, and results show a significant reduction in total harmonic distortion, improving current ripples compared to predictive control that does not use the multirate technique.