Machine learning models to predict electricity consumption and the impacts of COVID-19 in Portugal - Thesis

Aug 4, 2026

Upload by Maeve

This thesis analyzes how machine learning models and data from public sources can be used to predict electricity consumption in Portugal. Reliable forecasts are essential for efficient energy-sector management amid rising global energy demand. Portugal is a strong case because it relies heavily on energy imports and faces energy poverty. The study uses a data-driven approach to examine twelve years of consumption patterns and assess how the COVID-19 pandemic, climatic patterns, and GDP influence electricity demand. Five predictive models (SARIMA, SARIMAX, VAR, SVR, LSTM) are evaluated across stable and unstable periods, including and excluding COVID-19.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages80

Reading Time3 h 22 min

AccessFree to download

Resumo

Metodologia e dados

Modelos e avaliação

Limitações e implicações

Department of Quantitative Methods for Management and Economics/ Department of Information Science and Technology Machine learning models to predict electricity consumption and the impacts of COVID-19 in Portugal Ana Sofia Lobato Sucena Master in Data Science Supervisor: PhD, Anabela Ribeiro Dias da Costa, Assistant Professor ISCTE-IUL Supervisor: PhD, Diana Elisabeta Aldea Mendes, Associate Professor ISCTE-IUL Department of Quantitative Methods for Management and Economics/ Department of Information Science and Technology Machine learning models to predict electricity consumption and the impacts of COVID-19 in Portugal Ana Sofia Lobato Sucena Master in Data Science Supervisor: PhD, Anabela Ribeiro Dias da Costa, Assistant Professor ISCTE-IUL Supervisor: PhD, Diana Elisabeta Aldea Mendes, Associate Professor ISCTE-IUL Dedication to my Family, Teachers, and Friends. Acknowledgments I want to express my sincere gratitude to my parents because, without their unconditional support and love throughout my life, I would not be who I am today or have accomplished what I have. I'd also like to express my gratitude to my brother and friends, who have always been there forme no matter what. The professors Anabela Costa and Diana Aldea Mendes, my thesis supervisors, deserve a special thank you for their patience, understanding, and positive attitude throughout the entire process. iv Resumo Esta tese analisa a forma como os modelos de “machine learning” e os dados provenientes de fontes de dados públicas podem ser utilizados para prever o consumo de eletricidade em Portugal. Boas previsões são cruciais para uma gestão eficiente do setor energético, nomeadamente devido ao aumento da procura global de energia. Portugal apresenta um ótimo caso para a previsão de consumo, uma vez que depende significativamente de importações de energia e sofre de pobreza energética. O estudo utiliza uma metodologia baseada em dados para analisar doze anos de padrões deconsumo energético e analisar a forma como a pandemia do COVID-19, os padrões climáticos e o PIBafetam o consumo de eletricidade. Foram estudados cinco modelos preditivos-SARIMA, SARIMAX, VAR, SVR e LSTM – e os seus indicadores de desempenho em dois períodos diferentes ( um para os doze anos de analise, incluído durante o Covid-19, e outro apenas para dados antes do Covid-19) . Assim, este estudo permite avaliar a prestação dos modelos de machine learning em periodos estáveise não estáveis O estudo reconhece as suas limitações, como a falta de dados na era pós-COVID, mas continua afornecer informações úteis para o desenvolvimento e a gestão de políticas energéticas. Palavras-chave: Consumo de eletricidade, Pandemia COVID-19, Modelos preditivos vi

FAQs

What is the main goal of the thesis?

To evaluate how machine learning models and public data can predict electricity consumption in Portugal and quantify the impacts of COVID-19, climate patterns, and GDP.

Which predictive models are tested?

The thesis studies five models: SARIMA, SARIMAX, VAR, SVR, and LSTM.

How are the models evaluated across time periods?

Model performance is assessed in two different periods: one covering the full twelve-year analysis including COVID-19, and another using data only before COVID-19.

Aug 4, 2026

Upload by Maeve