| Name: | Description: | Size: | Format: | |
|---|---|---|---|---|
| 2.34 MB | Adobe PDF |
Authors
Advisor(s)
Abstract(s)
A presente dissertação tem por objetivo aplicar o modelo de previsão, Redes Neuronais Artificiais, aplicadas a séries de dados de consumo de energia elétrica em horizontes temporais de curto prazo.
Neste trabalho procura-se conhecer futuros valores de consumo de energia elétrica através de dados conhecidos relativos a acontecimentos passados, ou seja, conhecendo o comportamento da série de valores e seus respetivos parâmetros efetuar-se-á uma previsão
do que irá acontecer. Alterando os vários parâmetros das Redes Neuronais Artificiais se verificará qual estrutura terá um melhor desempenho na previsão do consumo de eletricidade em Portugal.
Os dados do passado relativo aos consumos verificados serão disponibilizados pela REN (Redes Energéticas Nacionais) e efetuar-se-á a análise do comportamento desta série temporal para posterior previsão da série temporal.
Como forma de validação da aplicabilidade das previsões serão calculados valores de erro absoluto, erro médio quadrático, erro percentual médio absoluto e intervalos de confiança para os valores a prever.
A metodologia aplicada nesta dissertação será a da previsão do consumo elétrico dos dias de semana e fins de semana para os períodos de 3,6,9 e 18 semanas de dados de entrada de 2008. Posteriormente avaliou-se a aplicação das Redes Neuronais Artificiais a outros períodos temporais de 2018 e analisou-se a sua eficiência.Para se conseguir uma boa gestão e produção de energia elétrica é cada vez mais importante ter um conhecimento prévio do comportamento futuro, o qual é impossível obter com exatidão.
This master’s thesis aims to apply the prediction methodology, Artificial Neural Networks, applied to electrical energy consumption data series in short-term time horizons. This work seeks to be able to predict the future values of electricity consumption through known data related to past events, that is, knowing the behavior of the series of values and their respective parameters will make a prediction of what will happen. Changing the various parameters of the Artificial Neural Network will verify which structure will have a better performance in forecasting electricity consumption in Portugal. The Portugal electrical past consumption data were available by REN (Redes Energéticas Nacionais) and an analysis of the behavior of this time series will be carried out for subsequent forecast of the time series. As a way of validating the applicability of the predictions, absolute error values, mean square error, mean absolute percentage error and confidence intervals for the values to be predicted will be calculated. The methodology applied in this dissertation will be the forecast of the electric consumption of the weekdays and weekends for the periods of 3, 6, 9, and 18 weeks of input data from 2008. Subsequently, it was analyzed the application of Artificial Neural Networks to other time periods in 2018. To achieve good electricity management and production, it is increasingly important to have prior knowledge of future behavior, which is impossible to obtain with exactness.
This master’s thesis aims to apply the prediction methodology, Artificial Neural Networks, applied to electrical energy consumption data series in short-term time horizons. This work seeks to be able to predict the future values of electricity consumption through known data related to past events, that is, knowing the behavior of the series of values and their respective parameters will make a prediction of what will happen. Changing the various parameters of the Artificial Neural Network will verify which structure will have a better performance in forecasting electricity consumption in Portugal. The Portugal electrical past consumption data were available by REN (Redes Energéticas Nacionais) and an analysis of the behavior of this time series will be carried out for subsequent forecast of the time series. As a way of validating the applicability of the predictions, absolute error values, mean square error, mean absolute percentage error and confidence intervals for the values to be predicted will be calculated. The methodology applied in this dissertation will be the forecast of the electric consumption of the weekdays and weekends for the periods of 3, 6, 9, and 18 weeks of input data from 2008. Subsequently, it was analyzed the application of Artificial Neural Networks to other time periods in 2018. To achieve good electricity management and production, it is increasingly important to have prior knowledge of future behavior, which is impossible to obtain with exactness.
Description
Trabalho Final de Mestrado para obtenção do grau de Mestre em Engenharia Eletrotécnica
Keywords
Previsão Série temporal Energia Elétrica Redes Neuronais Artificiais Forecasting Time series Electrical Energy Artificial Neural Networks
Citation
RODRIGUES, Fábio Flipe da Silva – Redes Neuronais Artificiais na Previsão do Consumo de Energia. Lisboa: Instituto Superior de Engenharia de Lisboa. 2023. Dissertação de Mestrado.
Publisher
Instituto Superior de Engenharia de Lisboa