Hyppää sisältöön
    • Suomeksi
    • På svenska
    • In English
  • Suomi
  • Svenska
  • English
  • Kirjaudu
Hakuohjeet
JavaScript is disabled for your browser. Some features of this site may not work without it.
Näytä viite 
  •   Ammattikorkeakoulut
  • Vaasan ammattikorkeakoulu
  • Opinnäytetyöt (Avoin kokoelma)
  • Näytä viite
  •   Ammattikorkeakoulut
  • Vaasan ammattikorkeakoulu
  • Opinnäytetyöt (Avoin kokoelma)
  • Näytä viite

Machine Learning and Statistical Approaches for Forecasting Electricity Demand in Vaasa

Karunarathne, Dinesh (2025)

Avaa tiedosto
Karunarathne_Dinesh.pdf (3.588Mt)
Lataukset: 


Karunarathne, Dinesh
2025
All rights reserved. This publication is copyrighted. You may download, display and print it for Your own personal use. Commercial use is prohibited.
Näytä kaikki kuvailutiedot
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:amk-2025060219444
Tiivistelmä
Electricity is vital to modern life, but stable supply is increasingly challenged by changing consumption patterns, climate variability, and renewable integration. This research focuses on forecasting electricity demand in Vaasa, Finland, a regional energy hub. Tailored short and long-term prediction models were developed to address grid stability challenges driven by factors such as climate variability.

The study utilized established load forecasting literature, comparing traditional statistical models against machine learning techniques and identifying key demand drivers. A quantitative methodology was employed using historical Vaasa-specific data, including electricity consumption and weather information. This involved comprehensive data acquisition, preprocessing, feature engineering (such as deriving 'Is_workday' and 'Sun_Flag' variables), and the iterative development and evaluation of Multiple Linear Regression (MLR), Random Forest, and XGBoost models for monthly, daily, and hourly forecasts, using metrics like R2 and MAPE.

The results demonstrated strong predictive capabilities, with an MLR model using temperature variables proving best for monthly forecasts with R2=96.3% (Table 6). For daily forecasts, MLR outperformed the tested machine learning models with R2=92.8% (Table 7). In hourly forecasting, XGBoost achieved a marginally better R2=90.2% (Table 17), closely followed by MLR with R2=90.1%. Temperature and various temporal features were consistently the most dominant demand drivers. The study concludes that for Vaasa's context, well-specified multiple linear models can be highly effective, offering valuable and interpretable tools for local energy management and strategic planning.
Kokoelmat
  • Opinnäytetyöt (Avoin kokoelma)
Ammattikorkeakoulujen opinnäytetyöt ja julkaisut
Yhteydenotto | Tietoa käyttöoikeuksista | Tietosuojailmoitus | Saavutettavuusseloste
 

Selaa kokoelmaa

NimekkeetTekijätJulkaisuajatKoulutusalatAsiasanatUusimmatKokoelmat

Henkilökunnalle

Ammattikorkeakoulujen opinnäytetyöt ja julkaisut
Yhteydenotto | Tietoa käyttöoikeuksista | Tietosuojailmoitus | Saavutettavuusseloste