| dc.description.abstract | Since the recent rise of European electricity prices, the field of electricity price forecasting (EPF) has gained increased popularity. With the renewable energy transition, the complexity of EPF has become more challenging due to the highly volatile nature of renewable energy sources. Additionally, the limited data covering the period with abnormally high prices, make EPF even more daunting. Accurate forecasts are therefore crucial in order to efficiently allocate energy resources. In this thesis, we introduce and investigate a novel approach to reduce the complexity of day-ahead EPF and better understand market coupling. Unlike univariate EPF, where features such as load, demand and weather forecasts are counted for, our approach is strictly based on multivariate electricity price time series from European electricity markets. We utilized an Long Short-Term Memory (LSTM) that successfully was capable of explaining electricity prices with varying accuracy throughout Europe. We found that the electricity market of Norway 1 (NO1) was the simplest to forecast, whereas the electricity markets of Denmark and Netherlands were the most difficult. Our LSTM yielded promising results and significantly outperformed a benchmark model using the same modeling approach. Nevertheless, multivariate price time series for EPF cannot be seen as the superior approach as its forecasts were lackluster in comparison to its counterpart. Moreover, using Local Interpretable Model Agnostic Explanations (LIME) we were able to quantify the importance of European electricity markets and analyze their interconnectivity. As expected, our results show that Germany and Great Britain are among the most influential. However, the electricity markets of Serbia and Croatia appear to have a strong connection with the high electricity prices. | en_US |
