Demonstrating the power of ML in predicting household appliance use
A UK-based research institution specialising in energy systems.
A UK energy research and demonstration body was exploring the concept of Home Energy Management Systems (HEMS). It wanted to investigate whether machine learning algorithms can learn future patterns of occupancy and needs of its residents using non-intrusive load monitoring (NILM) from mult-vector meter data and derived features.
We used advanced unsupervised machine learning to identify the unique energy signatures of different appliances using high resolution smart meter data. Our techniques were able to successfully plot when certain appliances were used without any prior knowledge of the appliances used or the household demographics. We were able to forecast whether a house was occupied and the use of appliances with a high degree of accuracy.
The results of the work suggest that both occupancy and appliance usage are predictable. This could pave the way for HEMS features that could maintain comfort in the home without compromising on costs e.g. through forecasting occupancy.