Privacy-Preserving Energy Optimisation in Home Automation Systems

This work explores the application of ML for time series forecasting in home automation systems. Specifically, we focus on predicting solar power generation and household energy consumption while prioritising user privacy and computational efficiency. The goal is to help users optimise their energy use by providing insights on surplus energy produced within the home. Our research investigates existing methods for time series forecasting and Deep Learning on resource-constrained devices. We consider techniques like ARIMA, Prophet, NeuralProphet, and RNNs, with a focus on transfer learning to address limitations on computational resources. 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 to deep learning models when properly deployed with resource-efficient techniques.