A carregar...
Publicação
Desagregação de consumos de smart homes por tipologia
| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 2.82 MB | Adobe PDF |
Orientador(es)
Resumo(s)
O presente trabalho tem como tema a desagregação de consumos no setor doméstico e a sua aplicação ao projeto re:dy da EDP, cuja frequência de amostragem dos valores de consumo dos clientes é relativamente baixa. Os conceitos da desagregação de consumos e do projecto EDP re:dy são explicados no capítulo de contextualização. O capítulo de metodologia contém uma explicação para cada um dos métodos matemáticos que foram utilizados proeminentemente durante o estágio. As abordagens exploradas para resolução do problema podem ser essencialmente divididas em dois processos preditivos: o processo utilizado para prever o consumo de frigoríficos e máquinas e o processo desenvolvido especificamente para a estimação do consumo de aquecimento ambiente. O primeiro processo segue uma estrutura de Ensemble Learning contando com 7 algoritmos e 5 meta-algoritmos cujos desempenhos são comparados após a análise dos valores preditos para o consumo de frigoríficos e máquinas dos clientes. Antes da construção do processo, as amostras dos consumos das categorias de equipamento em questão foram sujeitas a uma análise exploratória para facilitar a escolha de algoritmos mais adequados. O conjunto de variáveis independentes (input do algoritmo) foi derivado da informação disponível para todos os clientes e processado através de uma análise em componentes principais. O processo preditivo para consumo de aquecimento ambiente foi desenvolvido estudando o impacto desta classe de equipamentos no consumo global dos clientes ao longo do ano. Ao contrário do primeiro processo, que é maioritariamente constituído por modelos estatísticos, este é um algoritmo empírico. Por fim, no capítulo de discussão, analisam-se as várias abordagens e possíveis direções futuras da sua aplicação ao projeto.
The present work is focused on household energy disaggregation and its application to EDP's re:dy project, that has a relatively low sampling frequency for clients' consumption values. The concepts of energy disaggregation and of the re:dy project are explained in a contextualization chapter. The methodology chapter contains explanations for each mathematical method that was prominently used throughout the internship. The approaches to the problem can be split into two predictive processes: the process used to predict the consumption of fridges and washers and the process specifically designed to estimate the consumption of heaters. The first process follows an Ensemble Learning framework, including 7 algorithms and 5 meta algorithms whose performances are compared after analyzing the predicted values for fridges' and washers' consumption. Before the construction of the predictive process, the consumptions of the said equipment categories were analyzed to help choose better suited algorithms. The set of input variables was derived from the information available for every client and processed through a principal component analysis. The predictive process for heater consumption was developed by studying the impact of this equipment class on the global consumption of the clients throughout the year. Unlike the first approach, that was mainly composed of statistical models, this one is an empirical algorithm. Finally, in the discussion chapter, both approaches are analyzed as well as the future of their application to the project.
The present work is focused on household energy disaggregation and its application to EDP's re:dy project, that has a relatively low sampling frequency for clients' consumption values. The concepts of energy disaggregation and of the re:dy project are explained in a contextualization chapter. The methodology chapter contains explanations for each mathematical method that was prominently used throughout the internship. The approaches to the problem can be split into two predictive processes: the process used to predict the consumption of fridges and washers and the process specifically designed to estimate the consumption of heaters. The first process follows an Ensemble Learning framework, including 7 algorithms and 5 meta algorithms whose performances are compared after analyzing the predicted values for fridges' and washers' consumption. Before the construction of the predictive process, the consumptions of the said equipment categories were analyzed to help choose better suited algorithms. The set of input variables was derived from the information available for every client and processed through a principal component analysis. The predictive process for heater consumption was developed by studying the impact of this equipment class on the global consumption of the clients throughout the year. Unlike the first approach, that was mainly composed of statistical models, this one is an empirical algorithm. Finally, in the discussion chapter, both approaches are analyzed as well as the future of their application to the project.
Descrição
Relatório de estágio de mestrado em Estatística e Investigação Operacional, apresentada à Universidade de Lisboa, através da Faculdade de Ciências, em 2018
Palavras-chave
Desagregação de consumos Baixa frequência EDP re:dy Ensemble Learning Data Science Teses de mestrado - 2018