Year: 2025
Type: dissertation
Abstract: The growing integration of renewable energy sources, such as wind and solar, has increased the importance of accurate electricity price forecasting. As the share of renewables rises, real-time balancing of supply and demand becomes more complex due to the intermittent and less predictable nature of these sources. Intraday markets offer a mechanism for market participants to adjust their positions after the day-ahead market closes, helping to manage imbalances closer to delivery. Accurate intraday price forecasts are therefore useful for market participants to optimize trading strategies, reduce imbalance costs and manage operational risk in increasingly dynamic electricity markets. This thesis focuses on forecasting prices in the Finnish intraday electricity market, specifically the ID3 price index, a key market reference. In order to evaluate forecasting effectiveness, two machine learning models, Long Short-Term Memory (LSTM) networks and Extreme Gradient Boosting (XGBoost), were compared to naive benchmark methods, including the latest trade price and the volume-weighted average price from the prior half hour. Feature analysis was carried out to identify factors influencing price formation, showing strong correlations between prior intraday prices and the ID3 price, as well as minor contributions from, for example, wind generation forecasts, production supply developments and electricity demand patterns. The study found that both LSTM and XGBoost models provided small improvements in forecasting accuracy over naive methods, although the naive models established a high baseline that was difficult to surpass. Importantly, LSTM and XGBoost were also capable of generating probabilistic forecasts, providing valuable uncertainty estimates that increased their practical utility. LSTM slightly outperformed XGBoost in both point and probabilistic forecasting, while XGBoost remained a strong and competitive alternative.
Source: Aaltodoc (Aalto University)
Author: Takaneva, Jonna
Language: English
Subfield: Electrical and Electronic Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7061994034