Universität zu Köln

Trebbien, Julius (2023). Explainable Artificial Intelligence and Deep Learning for Analysis and Forecasting of Complex Time Series: Applications to Electricity Prices. Masters thesis, Universität zu Köln.

PDF JTrebbien_Thesis_final.pdf Download (4MB)

Abstract

A rapid energy transition from fossil fuel based generation to renewable energy sources is vital for the mitigation of climate change but requires complex market structures to manage the coordination of generation and demand. In particular, the German day-ahead market reacts to short-term forecasts one day prior to delivery and is driven by various external drivers. Its understanding and forecasting are essential for the energy transition as it allows renewable energy operators to make profits and promotes key technologies for a stable grid operation, such as battery storage. In this work, we analyze the German day-ahead electricity market using eXplainable Artificial Intelligence (XAI) and forecast electricity prices using deep neural networks. We investigate the application of SHapley Additive exPlanations (SHAP) to study the driving factors of electricity prices. The dataset includes several power system features such as load or renewable forecasts but also fuel prices. Our analysis suggests that load, wind and solar generation are the central external features driving prices, as expected, wherein wind generation affects prices more than solar generation. Simi- larly, fuel prices also highly affect prices in a nontrivial manner. Moreover, large generation ramps are correlated with high prices due to the limited flexibility of nuclear and lignite plants. Based on the results from the XAI method, we establish Long Short-Term Memory (LSTM) networks to forecast electricity prices. We introduce a probabilistic forecast as output, increas- ing the applicability of the model. The LSTM model is able to outperform models from related works and enables additional applications using the predicted standard deviation.

Item Type:	Thesis (Masters thesis)
Creators:	Creators Email ORCID ORCID Put Code Trebbien, Julius juliustrebbien@gmail.com UNSPECIFIED UNSPECIFIED
URN:	urn:nbn:de:hbz:38-707668
Date:	15 March 2023
Language:	English
Faculty:	Faculty of Mathematics and Natural Sciences
Divisions:	Faculty of Mathematics and Natural Sciences > Department of Physics > Institute for Theoretical Physics
Subjects:	no entry
Uncontrolled Keywords:	Keywords Language Electricity Prices English Explainable Machine Learning English Deep Learning English
Date of oral exam:	24 April 2023
Referee:	Name Academic Title Witthaut, Dirk Prof. Dr. Rydin Gorjão, Leonardo Prof. Dr.
Refereed:	Yes
URI:	http://kups.ub.uni-koeln.de/id/eprint/70766