Autonomous model predictive control for realization of smart inverters at the grid-edge

Abstract

The continuous increase in feasibility of renewable energy and renewable energy assets has opened the door to a future grid dominated by renewable-based power electronics converters, rather than electric machines powered by fossil fuels. Simultaneously, increases in practical processor clock speed enables new, advanced control techniques for such devices. The sampling rate of digital controllers for power converters is now able to be set within the same order of magnitude as the converter’s power semiconductor switching frequency, and even greater. This has created the possibility for the controller to perform online decision-making. This is realized in a control technique called model predictive control, more specifically, finite-set model predictive control. In finite-set model predictive control, the control will evaluate each of its available control actions and select that which achieves best performance. This evaluation of possible control actions requires making a prediction about each control action based on an internal model of the controlled system. To define best performance, the control contains objectives and their associated reference value, or value which is considered optimal. The control action with predictions aligned most closely with the reference are selected as the next-up control action. The ability to individually evaluate and select potential control actions presents transient responses faster than can be seen in control systems that incorporate typical linear controllers and modulator-based switching. It also allows for single-loop multi-objective control and the possibility to program grid-tied inverters with enhanced system awareness, as the controller can note performance that results from its decided control action. In this thesis, I present finite-set multi-objective model predictive control for multiple grid-tied power electronics converters. I introduce practical enhancements to the finite-set model predictive control paradigm which can remove the controller design stage, a generally tedious and ambiguous for predictive controllers. I show how hierarchical objective tracking makes it possible to retain fast controller sampling rates on converter topologies with especially large control sets. Finally, I introduce solutions which enables real time model alignment, fault tolerant operation and situational awareness of the converter. These enhancements to predictive control can ensure that the smart inverters of tomorrow’s grid are fast, aware, and reliable.