Método de monitoramento não invasivo de cargas elétricas residenciais

Abstract

Non-invasive appliance load monitoring is a modern technique that has several applications bringing benefits for both consumers and electrical utilities. These techniques allow bill discrimination and detection of energy losses. It also provides valuable information for energy programs, and creates a better characterization of the loads. This work presents a new approach for non-invasive residential electrical load monitoring. The innovation of the proposed approach is the use of cumulants of second and fourth order extracted from the electric current signal of the residential electrical loads as signatures of these loads. In order to reduce the dimension problem, two methods for feature selection were employed: a) Fisher’s Linear Discriminant and Fisher’s Linear Discriminant combined to Genetic Algorithms. The selected features are presented to a classifier, which identifies the residential electric load class of the processed signal. Two different classifiers were used: a) Artificial Neural Networks and b) Decision Tree. Results from these approaches were comparatively presented. This work has considered eleven different classes of residential electrical loads. Results were obtrained out from experimental electric signals and a high performance is achieved.