Parts | Days | Selection | Search | Updates | Downloads | Help

SOE: Fachverband Physik sozio-ökonomischer Systeme

SOE 7: Poster

SOE 7.16: Poster

Monday, March 18, 2024, 18:00–20:30, Poster D

Transformer neural networks for the detection of artefacts in energy market data — •Henrike von Hülsen^1,2, Ulrich Oberhofer², Benjamin Schäfer², Oliver Lauwers³, and Gust Verbruggen⁴ — ¹50Hertz Transmissions GmbH, Berlin, Germany — ²Institute for Automation and Applied Informatics, Karlsruhe Institute of Technology, Eggenstein-Leopoldshafen, Germany — ³Elia Group, Brussels, Belgium — ⁴Microsoft, Brussels, Belgium

Participants in the Belgian electricity market are permitted to deviate from their scheduled production or consumption, if the deviation counteracts a current imbalance in the market. The interpretation of the emerging data on schedule deviations is disturbed by artefacts caused by ramping times of different assets.

Drawing inspiration from the success of transformer architectures in handling 1D imaging data, the hypothesis is that transformers can efficiently process the underlying time series data to identify and subsequently eliminate these artefacts. Removing the artefacts will directly contribute to the efficiency of the capacity market in Belgium.

We propose a method from the evolving filed of transformer applications in diverse data domains, that will reliably detect and remove ramping artefacts without diminishing the quality of the signal.

Keywords: Machine learning; Transformers; Artefact removal; Energy markets