Analysis of Day-Ahead Prices and Influencing Factors in the Nordic Power Market
Author
Nielsen, Anders Linnemann
Term
4. term
Education
Publication year
2023
Submitted on
2023-10-06
Abstract
This thesis evaluates how well different machine learning methods—Lasso (a linear model with built-in variable selection), neural networks, and random forests—forecast day-ahead electricity prices in the Nordic power market. Two benchmarks are used for comparison: ARIMA (a classical time-series model) and a naive model. The study compares 15 time series of day-ahead prices from 2015 through September 2023. To enhance the models, the machine learning approaches also use wind and solar power production forecasts as inputs. Performance is assessed with common error measures: MAE (Mean Absolute Error), MSE (Mean Squared Error), MAPE (Mean Absolute Percentage Error), and RMSE (Root Mean Squared Error). The results indicate that neural networks perform very well when tuned to avoid overfitting, and that Lasso provides stable results without extensive manual supervision.
Specialet undersøger, hvor godt forskellige maskinlæringsmetoder – Lasso (en lineær model med indbygget variabeludvælgelse), neurale netværk og random forests – kan forudsige dag-forud elpriser i det nordiske elmarked. Som reference bruges to benchmarkmodeller: ARIMA (en klassisk tidsrække-model) og en naiv model. Sammenligningen omfatter 15 tidsserier af dag-forud elpriser fra 2015 til og med september 2023. For at forbedre modellerne får maskinlæringsmetoderne også adgang til prognoser for vind- og solkraftproduktion. Ydeevnen vurderes med almindelige fejlkriterier: MAE (Mean Absolute Error), MSE (Mean Squared Error), MAPE (Mean Absolute Percentage Error) og RMSE (Root Mean Squared Error). Resultaterne viser bl.a., at neurale netværk kan klare sig meget godt, når de indstilles, så de undgår overtilpasning (overfitting), og at Lasso leverer stabile resultater uden omfattende manuel indgriben.
[This apstract has been rewritten with the help of AI based on the project's original abstract]