Secondary control activation analysed and predicted with explainable AI

The transition to a renewable energy system challenges power grid operation and stability. Secondary control is key in restoring the power system to its reference following a disturbance. Underestimating the necessary control capacity may require emergency measures, such that a solid understanding of its predictability and driving factors is needed. Here, we establish an explainable machine learning model for the analysis of secondary control power in Germany. Training gradient boosted trees, we obtain an accurate ex-post description of control activation. Our explainable model demonstrates the strong impact of external drivers such as forecasting errors and the generation mix, while daily patterns in the reserve activation play a minor role. Training a prototypical forecasting model, we identify forecast error estimates as crucial to improve predictability. Generally, input data and model training have to be carefully adapted to serve the different purposes of either ex-post analysis or forecasting and reserve sizing.