Essays on electricity price forecasting

• Ano de defesa: 2023
• Autor(a) principal: Tiago Silveira Gontijo
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Instituição de defesa: Universidade Federal de Minas Gerais; Brasil; ENG - DEPARTAMENTO DE ENGENHARIA PRODUÇÃO; Programa de Pós-Graduação em Engenharia de Produção; UFMG
• Palavras-chave em Português: Applied statistics; Big data; Data science; Exploratory data analysis; Electricity price; Machine learning; Quantitative methods; Systematic literature review; Time series; Forecasting; Engenharia de produção; Aprendizado do computador; Estatística aplicada; Pesquisa quantitativa; Energia elétrica - Preços; Previsão; Series temporais
• Link de acesso: http://hdl.handle.net/1843/52319; https://orcid.org/0000-0003-2636-899X

Developing predictive models is a complex task since it deals with the uncertainty and the stochastic behavior of variables. Specifically concerning commodities, accurately predicting their future prices allows for risk minimization and establishment of more reliable decision support mechanisms. Discussion of this issue is extensive, and academic attention is being paid to the construction of nonparametric models to be applied to energy markets. They have presented promising predictive results, which justifies this research. Given the above, the following question is formulated: How is it possible to predict energy prices accurately in the Brazilian spot market? The present thesis provides a systematic literature review of the main forecasting methods applied to the energy sector. In the present study, it was possible to identify research gaps and, thus, propose new predictive models. The present thesis presents predictive models based on the idea of analogs. Analogs consist of scanning a time series and identifying patterns (so-called "matches") that are similar to the last available observations. Additionally, the recent hierarchical time series prediction theory has been incorporated, since many energy databases have well-defined dependency patterns.