Non-Intrusive Load Monitoring of Single and Aggregated Profiles with a Hidden Markov Model

International audienceAwareness raising programs to encourage energy efficient behaviour is important in the context of the current energy transition. Sensors and connected devices allowing for data collection are easily available providing an opportunity to collect electric consumption data from individual appliances. The effective use of these data sources is necessary to optimize an energy efficiency coaching program. This paper presents a methodology for non-intrusive load monitoring analysis on individual smart plugs to identify an unknown appliance and disaggregated an aggregated load profile such as a power strip. The automatic detection of a connected appliance allows for appliance usage suggestions to be provided quickly without the need of an end-user input. The disaggregation of an aggregated curve such as a power strip allows for the optimization of the number of smart plugs required to represent a significant proportion of the total energy use of the household. The down sampling of high resolution data was also performed to observe the performance of the methodology on lower resolution data. The single appliance identification models all had very high accuracy (between 94-100 %). The disaggregation of an aggregated profile in the kitchen use case also had high accuracy for data with a resolution of less than one minute (95-99 %). The disaggregation of an aggregated profile for a multi-media use case had a lower performance when more than two appliances were considered (55-85 %)