Publicação
Smart and private domotics
| Resumo: | As energy prices soar, intelligent, privacy-focused solutions to optimize home energy consumption are increasingly crucial. This work explores the application of Machine Learning models for time series forecasting in home automation systems. We focus on predicting solar power generation and energy consumption within a household, prioritizing user privacy and computational efficiency for deployment in embedded systems. The goal is to help users optimize their energy usage by providing insights on the amount of surplus energy produced within the home. This research delves into existing methods for time series forecasting and the application of Deep Learning models on resource-constrained devices. Models such as ARIMA, Prophet, NeuralProphet, and Recurrent Neural Networks architectures were considered with a focus on Transfer Learning to address the limitations on computational resources of embedded systems. The results demonstrate that traditional statistical methods can be effective for specific tasks. Furthermore, while embedded systems may struggle to train models from scratch, they can achieve similar performance with Deep Learning models when properly deployed with resource-efficient techniques. |
|---|---|
| Autores principais: | Sousa, Henrique Carvalho |
| Assunto: | Time series forecasting Embedded systems Transfer learning Smart home User privacy Energy consumption optimization |
| Ano: | 2024 |
| País: | Portugal |
| Tipo de documento: | dissertação de mestrado |
| Tipo de acesso: | acesso aberto |
| Instituição associada: | Universidade de Aveiro |
| Idioma: | inglês |
| Origem: | RIA - Repositório Institucional da Universidade de Aveiro |
| Resumo: | As energy prices soar, intelligent, privacy-focused solutions to optimize home energy consumption are increasingly crucial. This work explores the application of Machine Learning models for time series forecasting in home automation systems. We focus on predicting solar power generation and energy consumption within a household, prioritizing user privacy and computational efficiency for deployment in embedded systems. The goal is to help users optimize their energy usage by providing insights on the amount of surplus energy produced within the home. This research delves into existing methods for time series forecasting and the application of Deep Learning models on resource-constrained devices. Models such as ARIMA, Prophet, NeuralProphet, and Recurrent Neural Networks architectures were considered with a focus on Transfer Learning to address the limitations on computational resources of embedded systems. The results demonstrate that traditional statistical methods can be effective for specific tasks. Furthermore, while embedded systems may struggle to train models from scratch, they can achieve similar performance with Deep Learning models when properly deployed with resource-efficient techniques. |
|---|