Publicação
A recommender system for energy-efficient strategies considering solar panel systems
| Resumo: | The increasing global focus on sustainability and energy conservation has led to a rise in the development of smart technologies aimed at optimizing energy consumption in buildings. This thesis presents the design and implementation of a recommender system to provide optimal energy-saving strategies for occupants of buildings equipped with solar panels and energy storage systems. The system collects data on energy consumption, weather, energy production, electricity prices, and user preferences. Using machine learning algorithms, it recommends future actions that reduce the energy footprint while maintaining user comfort. This thesis also evaluates the system’s effectiveness and its potential impact on energy conservation. As a dataset containing all these components was not readily available, the first step was to create a new dataset encompassing all the relevant areas. This was done by gathering data from online databases or directly from users, resulting in a dataset of thirteen households for analysis. A data preparation pipeline was developed for the recommender system. Inputs included predictions of energy consumption, weather conditions, energy production, and electricity prices. After reviewing common models used for these predictions, various machine-learning techniques were tested to determine the best-fit model. The techniques tested included LSTMs, RFR, MLP, XGBoost, and SVR for regression, as well as RF, DT, KNN, SVC, GNB, MLP, SGD, and XGBoost for classification. With the final predictions, heuristic analysis was applied to determine the optimal energy source at each moment. A collaborative filtering approach was then used to enhance user recommendations based on days. The project demonstrated significant monthly price reductions, ranging from 20% to 70%, proving its effectiveness in providing energy-efficient strategies. |
|---|---|
| Autores principais: | Rodrigues, Tiago Pereira |
| Assunto: | Eficiência Energética Sistema de Recomendação Otimização de Custos Modelação Preditiva Painéis Solares Teses de mestrado - 2024 |
| Ano: | 2024 |
| País: | Portugal |
| Tipo de documento: | dissertação de mestrado |
| Tipo de acesso: | acesso aberto |
| Instituição associada: | Universidade de Lisboa |
| Idioma: | inglês |
| Origem: | Repositório da Universidade de Lisboa |
| Resumo: | The increasing global focus on sustainability and energy conservation has led to a rise in the development of smart technologies aimed at optimizing energy consumption in buildings. This thesis presents the design and implementation of a recommender system to provide optimal energy-saving strategies for occupants of buildings equipped with solar panels and energy storage systems. The system collects data on energy consumption, weather, energy production, electricity prices, and user preferences. Using machine learning algorithms, it recommends future actions that reduce the energy footprint while maintaining user comfort. This thesis also evaluates the system’s effectiveness and its potential impact on energy conservation. As a dataset containing all these components was not readily available, the first step was to create a new dataset encompassing all the relevant areas. This was done by gathering data from online databases or directly from users, resulting in a dataset of thirteen households for analysis. A data preparation pipeline was developed for the recommender system. Inputs included predictions of energy consumption, weather conditions, energy production, and electricity prices. After reviewing common models used for these predictions, various machine-learning techniques were tested to determine the best-fit model. The techniques tested included LSTMs, RFR, MLP, XGBoost, and SVR for regression, as well as RF, DT, KNN, SVC, GNB, MLP, SGD, and XGBoost for classification. With the final predictions, heuristic analysis was applied to determine the optimal energy source at each moment. A collaborative filtering approach was then used to enhance user recommendations based on days. The project demonstrated significant monthly price reductions, ranging from 20% to 70%, proving its effectiveness in providing energy-efficient strategies. |
|---|