A carregar...
Publicação
Previsão de preços de eletricidade no mercado spot baseado em Machine Learning
| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 4.53 MB | Adobe PDF |
Autores
Resumo(s)
A transição energética na Europa e a crescente integração de fontes renováveis têm
intensificado a volatilidade dos preços no mercado de eletricidade da Península Ibérica. Este
cenário apresenta desafios para o equilíbrio entre a oferta e a procura, agravados pela
intermitência associada à integração de energias renováveis. O problema central desta
dissertação é desenvolver um sistema avançado de previsão de preços no mercado spot, que
permita mitigar os impactos desta volatilidade, promovendo estabilidade e eficiência. Para
atingir este objetivo, foi seguido um planeamento estruturado segundo a metodologia Action
Research, incluindo uma revisão da literatura com PRISMA e a estruturação do processo de
desenvolvimento com CRISP-DM. Esta abordagem permitiu identificar lacunas e avanços na
aplicação de modelos de Machine Learning na transição energética e nos mercados de energia
elétrica. Foram estudados e comparados modelos lineares, modelos baseados em árvores,
modelos de gradient boosting e redes neuronais recorrentes. Os resultados evidenciaram a
superioridade dos métodos de boosting na redução do erro absoluto, com o XGBoost a obter o
melhor desempenho, seguido dos modelos de ensemble, superando as RNNs. As principais
limitações encontradas relacionam-se com a janela temporal, a parametrização uniforme nas
RNNs e a manutenção de outliers.
The energy transition in Europe and the growing integration of renewable sources have intensified price volatility in the Iberian Peninsula electricity market. This scenario presents challenges for balancing supply and demand, exacerbated by the intermittency associated with the integration of renewable energies. The central problem of this dissertation is to develop an advanced spot market price forecasting system that mitigates the impacts of this volatility, promoting stability and efficiency. To achieve this goal, a structured plan was followed according to the Action Research methodology, including a literature review with PRISMA and the structuring of the development process with CRISP-DM. This approach allowed us to identify gaps and advances in the application of Machine Learning models in the energy transition and electricity markets. Linear models, tree-based models, gradient boosting models, and recurrent neural networks were studied and compared. The results showed the superiority of boosting methods in reducing absolute error, with XGBoost achieving the best performance, followed by ensemble models, surpassing RNNs. The main limitations encountered relate to the time window, uniform parameterization in RNNs, and the maintenance of outliers.
The energy transition in Europe and the growing integration of renewable sources have intensified price volatility in the Iberian Peninsula electricity market. This scenario presents challenges for balancing supply and demand, exacerbated by the intermittency associated with the integration of renewable energies. The central problem of this dissertation is to develop an advanced spot market price forecasting system that mitigates the impacts of this volatility, promoting stability and efficiency. To achieve this goal, a structured plan was followed according to the Action Research methodology, including a literature review with PRISMA and the structuring of the development process with CRISP-DM. This approach allowed us to identify gaps and advances in the application of Machine Learning models in the energy transition and electricity markets. Linear models, tree-based models, gradient boosting models, and recurrent neural networks were studied and compared. The results showed the superiority of boosting methods in reducing absolute error, with XGBoost achieving the best performance, followed by ensemble models, surpassing RNNs. The main limitations encountered relate to the time window, uniform parameterization in RNNs, and the maintenance of outliers.
Descrição
Palavras-chave
Electricity market Energy transition Volatility Forecasting models Machine Learning Mercado de Eletricidade Transição energética Volatilidade Modelos de previsão