Data-driven control of DC-DC power converters

In this thesis we develop a data-driven approach to the control design of power convert- ers. We show that, given a set of measured data containing information about variables of interest (duty cycle as the input and inductor current/capacitor voltage as outputs) in the system, we can achieve asymptotic stability by solving a set of Lyapunov linear matrix inequalities (LMIs). This approach effectively addresses the issue of performance degradation in controllers operating over networks, i.e. feeding constant power loads (CPL) as opposed to their standalone design, i.e. with nominal resistive loads. In order to do so, we study elements of behavioral system theory such as linear difference systems and quadratic difference forms; this allows for the creation of a framework compatible with higher-order discrete systems, which guarantees asymptotic stability in power converters both in standalone operation and with increased modeling complexity when interconnected to a network. Moreover, given the fact that the aforementioned LMIs provide us with multiple stabilizing gain solutions, we develop an algorithm for the synthesis of a switching multi- controller framework which, given a family of controllers, endeavors to select the single best-performing set in order to improve the dynamic profile of a to-be-controlled system, e.g. a power converter. Simulations and experimental results are provided as proof of concept, thus validating the theoretical material and illustrating the advantages of the proposed approaches.