Application of a Deep Learning Generative Model to Load Disaggregation for Industrial Machinery Power Consumption Monitoring

Abstract

Non-Intrusive Load Monitoring (NILM) or Load Disaggregation is a set of techniques to identify and monitor loads from readings of aggregated signals from a unique electricity meter on a building. This paper presents a new dataset of industrial electric energy consumption and compares Factorial Hidden Markov Model and a Deep Learning-based model to disaggregate six different industrial machines from a site meter on a factory in Brazil. The Deep Learning-based model reduced normalized disaggregation error (NDE) and signal aggregated error (SAE) in comparison with the FHMM models for the same appliances. It also increased percentage of time during which the machine is correctly classified as turned ON or OFF.