Neuro-Dynamic State Estimation of Microgrids

The increasing integration of renewable energy brings complicated dynamics in networked microgrids, calling for high-fidelity dynamic state estimation of those inverter-based resources. Traditional dynamic state estimation, which requires accurate physical models of each individual inverter and the entire networked microgrids, is becoming increasingly unattainable. This chapter presents neuro-dynamic state estimation, a learning-based dynamic state estimation algorithm for inverter-interfaced networked microgrids under unknown subsystems. Wefirst establish a data-driven neuro-dynamic state estimation algorithm for networked microgrids with partially unidentified dynamic models, which incorporates the neural-ordinary-differential-equations into Kalman filters. We further develop a self-refining neuro-dynamic state estimation algorithm that enables data-driven dynamic state estimation under limited and noisy measurements by establishing automatic filtering, augmenting, and correcting framework.