A Gaussian process framework for the probabilistic dynamic modeling of active distribution networks using exogenous variables

The stochastic nature of load and distributed generation (DG) poses many challenges in the development of realistic dynamic models for active distribution networks (ADNs). The consumers constantly alter their power demand both in magnitude and composition while the DG output is governed by weather conditions. To this end, we introduce a dynamic ADN model that can directly incorporate time and weather variables in order to decode their influence and eventually yield more accurate predictions. To do so, we tailor a Gaussian process (GP) model such that it can capture the nonlinear ADN dynamics while learning any relevant information encoded in a list of exogenous variables such as time of the day, month, air temperature, etc. Furthermore, in addition to a traditional dynamic response, the model generates the associated predictive uncertainty. Finally, the ability of the proposed model to reflect the time-varying and weather dependent dynamics of load and DG is highlighted using field data acquired over a year in four different substations spread out in Southern Germany.