Model Predictive Direct Current Control for multi-level converters

A model predictive current controller for multi-level inverter driving electrical machines is proposed that keeps the stator currents within given bounds around their respective references and balances the inverter's neutral point potential around zero. The inverter switch positions are directly set by the controller thus avoiding the use of a modulator. Admissible switching sequences are enumerated and a state-space model of the drive is used to predict the drive's response to each sequence. The predicted short-term switching losses are evaluated and minimized. The concept of extrapolation and the use of bounds achieve an effective prediction horizon of up to 100 time-steps despite the short switching horizon. When compared to classic modulation schemes such as pulse width modulation, for long prediction horizons, the switching losses and/or the harmonic distortion of the current are almost halved when operating at low pulse numbers, thus effectively resembling the steady-state performance of optimized pulse patterns. During transients the dynamic response time of the proposed controller is in the range of a few ms and thus very fast.