A Contrastive NILM Approach for Appliance Detection

Detecting the power consumption of individual appliances is crucial for minimizing overall power usage. One effective method to achieve this is by identifying the on and off events of each appliance. This article investigates the problem of identifying individual electrical loads in a house by analyzing these events. It proposes a system that utilizes a nonintrusive load monitoring (NILM) technique to extract the energy demand of each device and generate spectrograms. Moreover, through the collected spectrograms, the representation encoder learns the representations by using a supervised contrastive learning loss, thereby enhancing the final classification of events.