Physics-Informed Learning Framework for Model-Free Controller Identification of Black/Gray-box DC-AC Converters

Obtaining accurate controller information in terms of control structures and parameters would be unrealistic for real functioning power electronic converter-based systems (PECs), which hinders the stability analysis of the PECs via an analytical model. This paper proposes a data-driven nonlinear controller identification method to shed insight into the gray/black box model through outside signal acquisitions. It contains mainly two stages. Firstly, a tailor-made library of candidate dynamics, containing possibly existing control structures, is devised systematically as prior knowledge. Secondly, a sparse regression algorithm is reformulated in order to extract the relevant structures in the library while obtaining controller parameters. This approach demonstrates that the sparse vector between the tailor-made library and output signal is closely related to the controller structure so that the PEGs can be governed by the identification model. The validity and correctness of the proposed method are verified by a 10-kVA experimental platform.