| Name: | Description: | Size: | Format: | |
|---|---|---|---|---|
| 5.57 MB | Adobe PDF |
Abstract(s)
Devido ao contínuo aumento do consumo de eletricidade e a possibilidade do esgotamento de recursos não renováveis, surge uma necessidade de uma melhor gestão do setor energético. Assim, com a inserção de fontes renováveis de energia na rede elétrica, uma das tarefas mais importantes revela-se ser a previsão do consumo de energia. Como tal, a obtenção de uma previsão do consumo de energia confiável é essencial para o controlo do consumo e a manutenção do equilíbrio nas redes de distribuição. Esta tese apresenta uma comparação entre soluções de Machine Learning tradicionais e Deep Learning para o problema da previsão do consumo de energia quando aplicadas a diferentes casos. Estas abordagens são apoiadas pela utilização da plataforma TensorFlow como fonte dos algoritmos utilizados. O objetivo deste estudo é explorar se o potencial fornecido pelo Deep Learning é capaz de superar a previsão de consumo dos métodos tradicionais, e fazê-lo quase em tempo-real. Os resultados mostram que apesar do Deep Learning ter capacidade para conseguir uma melhor precisão de previsão, necessita de grandes quantidades de dados como suporte.
Due to the increasing amount of today's electricity consumption and the possibility of the depletion of non-renewable resources, a necessity arose for better management of the energy sector. As such, with the insertion of renewable sources of energy into the grid, one of the most important tasks became the prediction of the energy consumption. This way, having a trustable forecast of the energy consumption is essential to control the consumption and maintain the balance in energy distribution networks. This thesis presents a comparison between traditional Machine Learning and Deep Learning solutions for the energy consumption forecasting problem, when applied to different scenarios. These approaches are supported by the usage of the TensorFlow platform as source for the algorithms used. The goal of this study is to explore whether the potential that Deep Learning provides can be used to surpass the forecasting of traditional Machine Learning, while doing so in near real-time execution. The results show that while Deep Learning can provide better precision of forecasting, it needs great amounts of data to feed and support it.
Due to the increasing amount of today's electricity consumption and the possibility of the depletion of non-renewable resources, a necessity arose for better management of the energy sector. As such, with the insertion of renewable sources of energy into the grid, one of the most important tasks became the prediction of the energy consumption. This way, having a trustable forecast of the energy consumption is essential to control the consumption and maintain the balance in energy distribution networks. This thesis presents a comparison between traditional Machine Learning and Deep Learning solutions for the energy consumption forecasting problem, when applied to different scenarios. These approaches are supported by the usage of the TensorFlow platform as source for the algorithms used. The goal of this study is to explore whether the potential that Deep Learning provides can be used to surpass the forecasting of traditional Machine Learning, while doing so in near real-time execution. The results show that while Deep Learning can provide better precision of forecasting, it needs great amounts of data to feed and support it.
Description
Keywords
Deep Learing Machine Learning TensorFlow Forecasting Energy Consumption Smart Grid