Household Energy Prediction: Methods and Applications for Smarter Grid Design

In this paper, we explore methods of generating accurate, real-time household energy usage predictions and the practical use cases for this prediction data. The ability to perform real-time prediction and the usefulness of such predictions are recent developments as connected smart energy devices become increasingly prevalent. These devices not only gather relevant data to learn historic trends, but can also improve overall grid functionality through direct device responsiveness. Machine learning has not yet been widely explored as an approach for this type of non-aggregated prediction, but we demonstrate its effectiveness as a tool even for this highly noisy data relative to other baseline and statistical approaches, and how all these approaches can complement each other. These predictions are crucial for enabling smart grid systems to effectively communicate their needs to the grid, and for the grid to appropriately prepare for future demand.