A carregar...
Publicação
Data-driven methodology for estimation of cruise ships and port energy profiles in small data settings
| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 7.15 MB | Adobe PDF |
Autores
Orientador(es)
Resumo(s)
Os portos são essenciais para o comércio global, mas enfrentam uma pressão crescente para cumprir
objetivos de sustentabilidade, em particular para reduzir as emissões de gases com efeito de estufa.
Uma parte importante destas emissões provém dos navios de cruzeiro, que necessitam de grandes
quantidades de energia enquanto atracados para manter serviços como HVAC, iluminação e in stalações para os passageiros. Esta necessidade de energia não propulsiva, conhecida como carga
hoteleira, depende tradicionalmente de geradores a bordo a combustíveis fósseis. Consequente mente, estimar com precisão esta necessidade energética é um passo crítico para a descarbonização
portuária e a transição para a eletrificação em larga escala.
Esta tese apresenta uma ferramenta baseada em modelos de machine learning para gerar estima tivas de consumo energético, incluindo intervalos de confiança, para navios de cruzeiro enquanto
atracados. Foram aplicadas e avaliadas várias técnicas de regressão, desde modelos lineares e não
lineares até abordagens probabilísticas, comparadas através de métricas de desempenho como o
erro quadrático médio e o coeficiente de determinação.
Os resultados mostram a viabilidade do uso de machine learning para prever a carga hoteleira,
apoiando o planeamento energético e estratégias de eletrificação como a shore side electricity. O
Porto do Funchal, na Ilha da Madeira, é o caso de estudo utilizado para validar esta metodologia.
Ports are essential to global trade but face increasing pressure to meet sustainability goals, par ticularly to reduce greenhouse gas emissions. An important portion of these emissions comes from cruise ships, which require substantial amounts of energy while at berth to sustain services such as HVAC, lighting, and guest facilities. This non-propulsive power requirement, known as the hotel load, traditionally relies on onboard fossil-fuel generators. Consequently, accurately estimating this demand is a critical step for port decarbonization and the transition toward large-scale electrifica tion. This thesis presents a tool based on machine learning models designed to generate energy consump tion estimates, including confidence intervals, for cruise ships while docked. Specifically, various regression techniques were applied and evaluated, ranging from linear and nonlinear models to probabilistic approaches. These models were compared using performance metrics, such as mean squared error and the coefficient of determination, to identify the most accurate methods for pre dicting consumption. The results show the feasibility of using machine learning to forecast hotel load demand, providing a basis for energy planning and supporting electrification strategies like shore side electricity. The Port of Funchal in Madeira Island serves as the case study for validating this methodology.
Ports are essential to global trade but face increasing pressure to meet sustainability goals, par ticularly to reduce greenhouse gas emissions. An important portion of these emissions comes from cruise ships, which require substantial amounts of energy while at berth to sustain services such as HVAC, lighting, and guest facilities. This non-propulsive power requirement, known as the hotel load, traditionally relies on onboard fossil-fuel generators. Consequently, accurately estimating this demand is a critical step for port decarbonization and the transition toward large-scale electrifica tion. This thesis presents a tool based on machine learning models designed to generate energy consump tion estimates, including confidence intervals, for cruise ships while docked. Specifically, various regression techniques were applied and evaluated, ranging from linear and nonlinear models to probabilistic approaches. These models were compared using performance metrics, such as mean squared error and the coefficient of determination, to identify the most accurate methods for pre dicting consumption. The results show the feasibility of using machine learning to forecast hotel load demand, providing a basis for energy planning and supporting electrification strategies like shore side electricity. The Port of Funchal in Madeira Island serves as the case study for validating this methodology.
Descrição
Palavras-chave
Shore side electricity Navios de cruzeiro Machine learning Previsão de consumo de energia Modelos de regressão Cruise ships Energy forecasting Regression models Informatics Engineering . Faculdade de Ciências Exatas e da Engenharia