A1121
Title: Day-ahead probability forecasting for redispatch 2.0 measures
Authors: Alla Petukhina - HTW Berlin (Germany) [presenting]
Maria Basangova - HTW Berlin (Germany)
Vlad Bolovaneanu - Bucharest Academy of Economic Studies (Romania)
Alexandra Conda - The Bucharest University of Economic Studies (Romania)
Awdesch Melzer - Humboldt University zu Berlin (Germany)
Christina Erlwein-Sayer - University of Applied Sciences HTW Berlin (Germany)
Abstract: The purpose is to advance a data-driven, day-ahead forecasting model for assessing the probability, direction, and scale of electrical congestions within Germany's complex power grid. We build a two-stage model, with the first stage employing an XGBoost classifier to predict the probability and direction of congestion events, and the second stage using a regression task for redispatch load forecasting. Utilising state-of-the-art machine learning algorithms, including TSMixer (Time-Series Mixer), LLMtime and NBEATSx (Neural basis expansion analysis with exogenous variables), the model is specifically designed to operate on an hourly basis, thereby offering timely insights for grid management. The analysis uncovers compelling evidence that key exogenous variables, such as real-time meteorological conditions, electricity supply-demand indicators, and Brent oil price fluctuations, can be harnessed to make highly reliable predictions concerning grid congestion events. The economic feasibility of the forecast was also analysed. The model can potentially serve as a useful resource for transmission system operators (TSOs) and policymakers interested in grid management and cost mitigation efforts.
