Deep Learning YOLO like model Applied on Energy Disaggregation

Appliance Load Monitoring (ALM) is nowadays crucial for energy management solutions, allowing user to obtain appliance-specific energy consumption data that can be used on design an energy optimization plan. Non-Intrusive Load Monitoring (NILM), or energy disaggregation, is currently the most attractive approach to address this problem since it avoids deploying a smart meter for each appliance on the system, but instead aim to disaggregate a specific appliance consumption from the total aggregated load data of a single building. We present a novel model inspirated on the architecture of YOLO, that stand for You Only Look Once, which objective is to detect in a single forward pass where is an appliance activation. We show that the proposed model can learn the singular signatures of the target appliances. The model was applied on several real-world household energy data, and achieve state-of-the-art performance, improving the detection on unseen houses due to learn the appliance signature even from the same appliance of different brands and user behaviours.