Probabilistic Prediction of the Area Control Error Using Normalizing Flows

Balancing generation and load is a central challenge in power systems, particularly those with a high share of renewable generation. The area control error (ACE) quantifies the current power mismatch in a certain area of the power grid and thus provides a central input for balancing and control. Accurate forecasting of this quantity can facilitate rapid control actions and thus improve grid stability. In this contribution, we introduce a probabilistic forecasting model for the ACE using a deep generative neural network model called normalizing flow. Our model generates scenarios for every quarter hour of the day using conditional features such as the generation schedules. We demonstrate that the generative model outperforms elementary benchmark models.