Office Appliances Identification and Monitoring using Deep Leaning based Energy Disaggregation for Smart Buildings

Abstract

Analysis of electrical energy metering profiles has experienced a substantial increase of research activity in recent years. This smart metering is a tool for monitoring energy usage and users' behaviors as a prerequisite for substantial energy savings. Instead of having a sensor at each appliance, non-Intrusive Load Monitoring (NILM) provides a cheaper solution by disaggregating the load data from a single meter using digital signal processing. Different algorithms have been successfully applied to a variety of load scenarios. Load data for small office appliances is available in the BLOND data set (Building-Level Office eNvironment Dataset) such as laptops, computer monitors, etc. The potential energy saving of each small appliance cannot be neglected, particularly in large companies/institutes. In this paper, a recurrent neural network (RNN) with long-short term memory (LSTM) is designed, trained, and validated for NILM on small power office equipment provided in the BLOND data set. A comparison to combinatorial optimization and factorial hidden Markov models using five metrics for performance testing shows good results for the proposed RNN. Index Terms-non-Intrusive Load Monitoring (NILM), recurrent neural networks, energy disaggregation, smart metering, smart buildings.