kth.sePublications
Change search
CiteExportLink to record
Permanent link

Direct link
Cite
Citation style
  • apa
  • ieee
  • modern-language-association-8th-edition
  • vancouver
  • Other style
More styles
Language
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Other locale
More languages
Output format
  • html
  • text
  • asciidoc
  • rtf
Forecasting Day-ahead and Intraday Electricity Prices Using Machine Learning
KTH, School of Electrical Engineering and Computer Science (EECS).
KTH, School of Electrical Engineering and Computer Science (EECS).
2024 (English)Independent thesis Basic level (degree of Bachelor), 10 credits / 15 HE creditsStudent thesis
Abstract [en]

This project conducts a comparative analysis of forecasting models to evaluatetheir performance in predicting electricity prices across hourly data for the day-ahead andintraday electricity markets within the Swedish bidding area SE3. A Long Short-TermMemory (LSTM) neural network, a Neural Basis Expansion Analysis (N-BEATS) neuralnetwork and an ensemble of models constructed through the AutoGluon library areexamined. The models are using external data as past covariates regarding predicted andactual power system load, date-time information, air temperature, wind speeds and solarradiation to improve the generated forecasts. Each model is evaulated on two time periodsfor each market, with the 2015-2019 period characterized by relative stability and the 2019-2023 period being marked by heightened volatility and uncertainty attributed to ongoingglobal events. The N-BEATS model struggled with forecasting prices in the 2015-2019 periodbut found more success in predicting the more volatile prices of the 2019-2023 period. Boththe LSTM model and the AutoGluon model found success in forecasting the 2015-2019period, with the latter achieving a Weighted Mean Absolute Percentage Error (wMAPE) of10.74% for the day-ahead market. All models generally performed worse in the 2019-2023period compared to the 2015-2019 period, with the LSTM model performing the best with awMAPE of 38.20% for the same market.

Abstract [sv]

Detta projekt genomför en jämförande analys av prognosmodeller för attutvärdera deras prestanda för att förutsäga elpriser över timdata för elmarknaderna fördagen framåt och inom dagen inom det svenska budområdet SE3. Ett neuralt nätverk medlånga korttidsminnen (LSTM), ett neuralt nätverk med neural basisexpansionsanalys (N-BEATS) och en ensemble av modeller konstruerade genom AutoGluon-biblioteketundersöks. Modellerna använder externa data som tidigare kovariater avseende förutspåddoch faktisk kraftsystembelastning, datum-tid-information, lufttemperatur, vindhastigheter ochsolstrålning för att förbättra de genererade prognoserna. Varje modell utvärderas på tvåtidsperioder för varje marknad, där priserna för perioden 2015-2019 kännetecknas av relativstabilitet och priserna för perioden 2019-2023 har en ökad volatilitet och osäkerhet, vilket kantillskrivas pågående globala händelser. N-BEATS-modellen hade problem med att förutsepriserna under perioden 2015-2019 men fann större framgång med att förutsäga de mervolatila priserna under perioden 2019-2023. Både LSTM-modellen och AutoGluon-modellenlyckades väl med att prognostisera perioden 2015-2019, där den senare uppnådde ett vägtgenomsnittligt absolut procentfel (wMAPE) på 10,74% för day-ahead-marknaden. Allamodeller presterade generellt sämre under perioden 2019-2023 jämfört med perioden 2015-2019, med LSTM-modellen presterandes bäst med en wMAPE på 38,20% för sammamarknad.

Place, publisher, year, edition, pages
2024. , p. 255-270
Series
TRITA-EECS-EX ; 2024:155
Keywords [en]
Electricity price forecasting, deep learning, LSTM, N-BEATS, AutoGluon, time series
National Category
Electrical Engineering, Electronic Engineering, Information Engineering
Identifiers
URN: urn:nbn:se:kth:diva-358450OAI: oai:DiVA.org:kth-358450DiVA, id: diva2:1928777
Supervisors
Examiners
Projects
Kandidatexamensarbete i Elektroteknik 2024Available from: 2025-01-17 Created: 2025-01-17

Open Access in DiVA

fulltext(121150 kB)38 downloads
File information
File name FULLTEXT01.pdfFile size 121150 kBChecksum SHA-512
82342f1408fe2aae929f55e76f2a176a8521cf94c0fe100464225724a9b74ddff6f61a0fb1cf957b5d6400be30877b9d0aec28080253ae7c1e09793536e10217
Type fulltextMimetype application/pdf

By organisation
School of Electrical Engineering and Computer Science (EECS)
Electrical Engineering, Electronic Engineering, Information Engineering

Search outside of DiVA

GoogleGoogle Scholar
Total: 38 downloads
The number of downloads is the sum of all downloads of full texts. It may include eg previous versions that are now no longer available

urn-nbn

Altmetric score

urn-nbn
Total: 639 hits
CiteExportLink to record
Permanent link

Direct link
Cite
Citation style
  • apa
  • ieee
  • modern-language-association-8th-edition
  • vancouver
  • Other style
More styles
Language
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Other locale
More languages
Output format
  • html
  • text
  • asciidoc
  • rtf