Event Classification in Non-Intrusive Load Monitoring Using Convolutional Neural Network

The present paper addresses the event classification in the scope of non-intrusive load monitoring (NILM) through the use of Convolutional Neural Network (CNN), an artificial intelligence (AI) kind. Concepts of CNN and NILM are presented, as well as the methodology for the elaboration of signature appliances from the aggregate monitoring of energy for CNN training. A case study is performed considering 34 events of different devices to test the classification performed by CNN. At this, was possible to observe that the use of CNN is adequate for the event detection on the NILM approach, reaching 87% accuracy in the classification on the performed test. Through the tests carried out, it was observed that some confusions occurred in the classification process can be corrected through the improvement of the method of signature set formation, in order to reach a wider range of the signatures spectrum.