Contribuições ao desenvolvimento de técnicas de controle preditivo baseado em modelo para conversores conectados à rede

Abstract

This thesis document presents contributions to the project and development of model predictive control techniques for voltage-fed grid-tied converters. In summary, the proposed algorithms have a low computational burden, which allows for increasing the sampling frequency for digital implementation. As a result, it is possible to optimize the passive elements of the output filters, reducing the volume and cost of the system. In this way, the present document consists of three scientific papers, published in international journals. In Article 1, a model predictive control algorithm is proposed based on the implementation of a single vector in each sampling period. In this approach, the system model is represented in the output line-to-line voltages coordinate system, where the converter vectors assume integer values, allowing the search for the nearest vector by rounding the reference. In addition, constraints are included to ensure switching only between adjacent levels of the converter output phase voltages. In Article 2, an algorithm with the implementation of a switching sequence during a sampling period is presented. In this work, the solution of an unconstrained optimization problem is considered to identify the sector in the space vector diagram where the reference vector lies. For the vectors of the selected sector, their respective duty-cycles are calculated by the Karush-Kuhn-Tucker conditions for both linear and over-modulation regions. Article 3 presents techniques for dead-time compensation of the semiconductor devices in a model predictive control algorithm using a switching sequence, aiming to improve the performance in steady-state operation. To do this, the prediction of the inverter-side currents during the switching period is proposed to predict the occurrence of dead time and compensate for it appropriately. Real-time simulation results based on Controller Hardware-in-the-Loop are presented, comparing the proposed techniques with similar algorithms found in the literature. Finally, experimental results demonstrate the good performance of the proposed model predictive control algorithms and dead-time compensation techniques.