Comparison of short-term forecast of hourly day-ahead electricity prices in DK1 using different time series and linear regression models
Author
Zukauskaité, Giedré
Term
4. semester
Publication year
2022
Submitted on
2022-05-31
Pages
38
Abstract
This thesis examines short-term forecasting of hourly day-ahead electricity prices in DK1 by comparing time series models (autoregressive and moving average) with a linear regression model. Models are estimated using maximum likelihood and selected with the Akaike Information Criterion (AIC). The regression model includes fossil fuel prices and CO2 allowance prices as predictors. Three separate datasets are used for the hours 7:00, 13:00, and 21:00 to represent different daily load periods. Forecast accuracy is assessed by comparing predicted and actual prices, including RMSE. The findings indicate that off-load periods are best predicted by linear regression, while the peak-load period is best predicted by an AR(5) model. The study highlights practical relevance for electricity trading, such as earlier capacity booking on interconnectors, and outlines a technical momentum trading strategy guided by model-driven day-ahead price movements.
Denne afhandling undersøger kortsigtet prognose af timevise day-ahead-elpriser i DK1 ved at sammenligne tidsseriemodeller (autoregressive og moving average) med en lineær regressionsmodel. Modellerne estimeres ved maksimum likelihood og udvælges med Akaike Information Criterion (AIC). Regressionsmodellen inkluderer priser på fossile brændsler og CO2-kvoter som forklarende variable. Tre separate datasæt anvendes for timerne kl. 7, kl. 13 og kl. 21 for at afspejle forskellige belastningsperioder over døgnet. Prognosenes nøjagtighed vurderes ved at sammenligne forudsagte og faktiske priser, herunder RMSE. Resultaterne viser, at off-load perioder bedst forudsiges af lineær regression, mens peak-load perioden bedst forudsiges af en AR(5)-model. Studiet peger på praktiske anvendelser i elhandel, herunder tidlig kapacitetsbooking på interconnectorer, og skitserer en teknisk momentumstrategi baseret på modeldrevne day-ahead prisbevægelser.
[This apstract has been generated with the help of AI directly from the project full text]