Our technology enabled a research institution to accurately record the electricity usage of different types of appliances, and accurately forecast whether a house would be occupied. This allowed heating to be turned off while the occupier is out and turned on in advance of them coming home – optimising heating and energy consumption.
A UK-based research institution specialising in energy systems.
The institution wanted to carry out non-intrusive load monitoring to understand which appliances were being used in a house at a particular time. This would help to optimise heating and energy consumption.
We used data from high-resolution electricity meters installed in test properties to find out which appliances were being used and when. This required analysing huge volumes of unlabelled data. We applied advanced unsupervised machine learning (ML) techniques to convert unstructured data into discernible patterns. In this way, we could accurately record the unique electricity usage signatures of different types of appliance. This helped automatically separate the signatures (power consumption patterns) of different groups of appliances into clear clusters.
Not only did the analysis reveal historical human activity within the building, but it also served to predict future needs. We could plot when appliances would be used. From this, we could predict, an hour in advance and with a high degree of accuracy, whether the house would be occupied. This allowed heating to be turned off while the occupier is out and turned on in advance of them coming home.
Our model demonstrates the power of ML as applied to smart-home controllers. It opens the way to advanced home hub devices that can optimise the domestic heating schedule in a predictive way. Our model would also have application in elderly care supervision, and security systems.