2024

Sridhar Araavind; Karhunen Markku; Honkapuro Samuli; Ruiz Fredy

There are two primary methods for predicting electricity prices: forecasting and nowcasting. This study compares these approaches by employing various machine learning algorithms to forecast electricity prices. The nowcast algorithms are trained on data spanning from 2018 to 2021 and evaluated for the years 2022 and 2023, during which the energy system of Finland underwent significant changes, whereas the forecasting algorithms use the data for the previous 90 days to predict the next-day prices. Among nowcasting methods, Random Forest emerged as the top-performing algorithm, while the k Nearest Neighbor algorithm performed best in the forecasting approach. Despite achieving relatively low prediction errors, the predicted prices for 2022 and 2023 diverged notably from the actual prices. This discrepancy underscores the challenge of accurately predicting prices using current open data sources, particularly in scenarios involving significant alterations in the energy system. Consequently, the ability to anticipate price changes based on energy system transformations remains elusive, impacting research efforts focused on price prediction under future-specific circumstances.