| Name: | Description: | Size: | Format: | |
|---|---|---|---|---|
| 8.36 MB | Adobe PDF |
Abstract(s)
Dado o crescimento exponencial da população e, consequentemente, o aumento do consumo de energia elétrica, surge a necessidade de prever a procura, para garantir um abastecimento ininterrupto e estável.
Uma vez que a eletricidade é consumida ao mesmo ritmo que é produzida nas centrais, o uso da previsão de energia, tanto no lado da produção como do lado do consumo, é fulcral para se manter uma rede de transporte e distribuição estável.
No entanto, o consumo não é um valor linear, sendo afetado por variáveis externas como a temperatura, o dia da semana e o mês do ano. Um outro fator que introduz aleatoriedade no sistema, são as Fontes de Energia Renováveis (FER), devido à sua componente estocástica referente ao seu recurso primário.
Neste trabalho é apresentado um modelo Multilayer Perceptron de previsão a curto prazo (24h), definido como um tipo de Rede Neuronal, considerada como um método de Inteligência Artificial, que tomou por base os dados relativos aos consumos de energia, registados durante os períodos de 2014 a 2017.
Para a previsão em estudo, foram obtidos resultados de MAPE de 3,43%, 4,66% e RMSE de 2639MW, 1201MW, para o modelo MLP e naive respetivamente.
As conclusões retiradas deste trabalho evidenciam que, através do modelo de previsão utilizado, se consegue obter resultados precisos, com margens de erro reduzida. Os resultados permitiram aferir que este método se apresenta como um método fiável para o estudo da previsão do consumo de energia.
Given the exponential growth of the population and, consequently, the increase in electricity consumption, the demand needs to be forecasted in order to guarantee an uninterrupted and stable supply. Since electricity is consumed at the same rate as it is produced in the power plants, the use of power forecasting on both the production and consumption side is key to maintaining a stable transmission and distribution network. However, consumption is not a linear value and is affected by external variables such as temperature, day of the week and month of the year. Another factor that introduces randomness into the system is the Renewable Energy Sources (RES), due to its stochastic component referring to its primary resource. This work presents a Multilayer Perceptron model for short-term (24h) forecasting, defined as a type of Neural Network, considered as a method of Artificial Intelligence, which was based on data related to energy consumption, recorded during the periods from 2014 to 2017. For the prediction under study, MAPE results of 3,43%, 4,66% and RMSE of 2639MW, 1201MW, were obtained for the MLP model and naive respectively. The conclusions drawn from this work show that, through the forecast model used, it is possible to obtain accurate results, with reduced error margins. The results allow us to conclude that this method is a reliable method for the study of energy consumption forecasting.
Given the exponential growth of the population and, consequently, the increase in electricity consumption, the demand needs to be forecasted in order to guarantee an uninterrupted and stable supply. Since electricity is consumed at the same rate as it is produced in the power plants, the use of power forecasting on both the production and consumption side is key to maintaining a stable transmission and distribution network. However, consumption is not a linear value and is affected by external variables such as temperature, day of the week and month of the year. Another factor that introduces randomness into the system is the Renewable Energy Sources (RES), due to its stochastic component referring to its primary resource. This work presents a Multilayer Perceptron model for short-term (24h) forecasting, defined as a type of Neural Network, considered as a method of Artificial Intelligence, which was based on data related to energy consumption, recorded during the periods from 2014 to 2017. For the prediction under study, MAPE results of 3,43%, 4,66% and RMSE of 2639MW, 1201MW, were obtained for the MLP model and naive respectively. The conclusions drawn from this work show that, through the forecast model used, it is possible to obtain accurate results, with reduced error margins. The results allow us to conclude that this method is a reliable method for the study of energy consumption forecasting.
Description
Dissertação de natureza científica para obtenção do grau de Mestre em
Engenharia Eletrotécnica - Ramo: Energia
Keywords
Modelos de previsão Previsão de cargas Previsão no curto prazo Redes Neuronais Forecast Models Load forecasting Short-term forecast Neural Networks
Citation
INÁCIO, Diogo Alexandre Trindade Ribeiro – Aplicação de Redes Neuronais para Previsão de Cargas no horizonte do planeamento operacional. Lisboa: Instituto Superior de Engenharia de Lisboa, 2023. Dissertação de Mestrado.
Publisher
Instituto Superior de Engenharia de Lisboa