End-to-end NILM System Using High Frequency Data and Neural Networks
Improving energy efficiency is a necessity in the fight against climate change. Non Intrusive Load Monitoring (NILM) systems give important information about the household consumption that can be used by the electric utility or the end users. In this work the implementation of an end-to-end NILM system is presented, which comprises a custom high frequency meter and neural-network based algorithms. The present article presents a novel way to include high frequency information as input of neural network models by means of multivariate time series that include carefully selected features. Furthermore, it provides a detailed assessment of the generalization error and shows that this class of models generalize well to new instances of seen-in-training appliances. An evaluation database formed of measurements in two Uruguayan homes is collected and discussion on general unsupervised approaches is provided.