Probabilistic Forecasting of Intraday Electricity Prices: In the Nordic Regions Using Deep Distributional Time Series Models
Translated title
Probabilistic Forecasting of Intraday Electricity Prices
Authors
Pedersen, Louise Neema Krog ; Nielsen, Leonora
Term
4. term
Education
Publication year
2023
Submitted on
2023-06-02
Pages
85
Abstract
Dette projekt undersøger probabilistisk prognosticering af intradags elpriser i de nordiske budområder (DK1-DK2, NO1-NO5, SE1-SE4, FI) med udgangspunkt i metoderne fra Klein et al. (2023). Tilgangen kombinerer Bayesiansk inferens for at kvantificere usikkerhed, Echo State Networks (ESN) til at fange ikke-lineære tidsmæssige mønstre, og copulaer til at modellere afhængighed på tværs af områder. Der udvikles og anvendes to modeller på Nord Pool-data: en Gaussisk probabilistisk ESN til de marginale prisdynamikker og en copula-baseret model til den fælles fordeling, inklusive estimationsprocedurer, marginal kalibrering og probabilistiske forecasts. I anvendelsen vurderes tæthedsprognoser og prognosenøjagtighed. Undersøgelsen peger på, at metoderne er lovende til at forbedre intradags prisprognoser og usikkerhedsvurdering, men også at de har begrænsninger i at indfange ekstreme hændelser og haleadfærd, hvilket indikerer behov for yderligere modeludvikling og kalibrering.
This project investigates probabilistic forecasting of intraday electricity prices in the Nordic bidding zones (DK1-DK2, NO1-NO5, SE1-SE4, FI) using methods inspired by Klein et al. (2023). The approach combines Bayesian inference to quantify uncertainty, Echo State Networks (ESNs) to capture nonlinear temporal dynamics, and copulas to model cross-area dependence. Two models are developed and applied to Nord Pool data: a Gaussian probabilistic ESN for marginal price dynamics and a copula-based model for the joint distribution, including estimation procedures, marginal calibration, and probabilistic forecasting. The application evaluates density forecasts and forecast accuracy. Findings indicate that while these methods are promising for intraday price forecasting and uncertainty quantification, they have limitations in representing extreme events and tail behavior, highlighting the need for further model development and calibration.
[This summary has been generated with the help of AI directly from the project (PDF)]
Keywords
Documents