Modeling German Energy Market Hourly Profiles with a Focus on Variable Renewable Energy

Detta är en Magister-uppsats från Lunds universitet/Nationalekonomiska institutionen; Lunds universitet/Statistiska institutionen

Sammanfattning: This paper investigates the best methods for modeling hourly profiles in the German energy market for the period between 2018 and 2022. Modeling emphasized variable renewable energy (VRE) and included information on the level of energy production, oil price, COVID lockdowns, and historic hourly energy spot prices. Previous research on energy prices has focused on interpretable models; while investigations emphasizing predictive accuracy are sparse and sequestered in industry. This paper is intended to contribute to the understanding of which algorithms and what variables (endogenous and exogenous to the energy market) are best at decreasing the discrepancies between predicted and observed hourly electricity prices. Four different algorithms were investigated for modeling, linear regression, lasso regression, gradient boosted trees, and a feed forward neural network. Gradient boosted trees accounted for the most variation with an R-squared of 87.7% and promising results on periods of high volatility. Oil price and the share of electricity generated by solar and wind were found to substantially improve predictive accuracy, while COVID lockdowns were less important for prediction. The results from this paper can be used to improve hourly energy price prediction or for comparison by future researchers on different methods.

  HÄR KAN DU HÄMTA UPPSATSEN I FULLTEXT. (följ länken till nästa sida)