Probabilistic Dynamic Model of Active Distribution Networks Using Gaussian Processes
Georgios Mitrentsis, Hendrik Lens · 2021
The development of an accurate yet generic dynamic equivalent model for active distribution networks (ADNs) is a rather delicate task due to the high uncertainty characterizing the various grid components. To this end, we propose a probabilistic non-parametric dynamic model that, except for its response, additionally yields the corresponding predictive uncertainty. The proposed model leverages a Gaussian process (GP) in order to learn the complex nonlinear dynamics of an ADN using measurement data while factoring in prior knowledge about the system. Real measurement data acquired in five substations spread out in Southern Germany highlight the ability of the proposed method to accurately capture the underlying dynamics of an ADN while readily providing confidence levels for the predictions. Finally, the potential and the challenges of using GPs for modeling ADNs are discussed.