Model Predictive Control for Systems with Limited State Variables

Model predictive control are actively used in industry in high-power electric drives, in which it is permissible to use expensive programmable logic matrices that provide calculations with a prediction horizon of up to ten epochs and a forecast tree of several million variants. However, the computing power of microcontrollers is not enough to implement such systems in a low-power electric drive. The development of microcontrollers is following the path of increasing not only computing power, but also memory, which makes it possible to implement model predictive control systems in practice using lookup table methods for calculating control actions for all combinations of state variables and control actions. The paper shows the possibility of designing control with prediction for two types of electric drives—a DC electric drive and a switched reluctance electric drive—as objects with a small number of state variables, which allows one to fit the precalculated lookup tables of control actions in the memory of the microcontroller. Studies on mathematical models have shown that model predictive control with a finite control set and a single prediction horizon has disadvantages due to variable switching frequency, and control with a continuous control set implemented by the lookup table method allows one to perform control calculations immediately before the start of the next pulse width modulation period, reducing the algorithm execution time by hundreds of times.