Comparison of Neural Networks for High-Sampling Rate NILM Scenario

The common objective of techniques employed to identify the use of household appliances is related to energy efficiency and the reduction of energy consumption. In addition, through load monitoring it is possible to assess the degree of independence of tenants with minimal invasion of privacy and thus develop sustainable health systems capable of providing the required services remotely. Both approaches should initially deal with the load identification stage. For that purpose, this work presents three different solutions that take the events of the electrical current signal acquired at high frequency and process them for classification by using two different topologies of Artificial Neural Networks (ANN). The data of interest used as input for the ANN in the proposals are the normalized signal captured around the events, the images created by dividing that signal into sections and organizing them in a matrix, and the images coming from the Short Time Fourier Transform (STFT) of the signal around the event. The dataset BLUED is used to carry out the validation of the proposal, where some of the proposed architectures obtain an F1 score above 90 % for more than fifteen devices under classification.