Machine Learning Techniques for Regression in Energy Disaggregation
Résumé
Non-Intrusive Load Monitoring (NILM) or Energy disaggregation may be the holy grail of energy efficiency. The impact of energy disaggregation at the commercial level of home customers is the increased utility customer engagement and the reduced energy usage. The goal at this level is to itemize the consumer’s energy bill, analyze the energy usage and cost per household appliance and make personalized and prioritized energy savings recommendations. All these should be viable through a single sensor per household that monitors the total energy consumption and other related quantities. Energy disaggregation is a set of computational approaches for extracting end-use appliance level data from an aggregate energy signal without any plug-level sensors. In the present work, we used a smart meter designed by Meazon S.A. to monitor the energy consumption of a house for 70 days and use basic machine learning methods for regression. To this end, we use an extensive set of features to train our models apart from using only active power. Furthermore, we make comparisons with respect to accuracy and training time between Decision Tree, Random Forest and k-NN machine learning methods.