Desagregação de energia utilizando GSP (Graph Signal Processing)

Abstract

The use of computational devices eventually generated a large accumulation of data. This data ranges from purchase receipts to lawsuits. Therefore, there is a strong business and academic interest in what to do with such data in order to extract important information from them. Work related to the problem of energy disaggregation and non-intrusive load monitoring involves creating data sets to study the problem, creating devices for data collection, among others. The problem of power disaggregation is to collect data from the house overviews, which is where the electric power coming from the poles is distributed to the house, and from the individual appliances and, using algorithms, to get data that comes very close to the data collected from the appliances, ie given the energy data of a home, check which appliances are using those loads at a particular time of day. This problem is part of Non-intrusive Load Monitoring (NILM), which is a set of techniques that involve hardware and software for monitoring power so that energy can be better utilized by residents. The solution used involves an algorithm called GSP (Graph Signal Monitoring), which uses signal processing to construct graphs with signal data (in the case of energy breakdown, the signals are electrical energy) and from these graphs, check which data are most similar and thus group them together to form the appliances that have consumed the electricity of the house. This method is a way of disaggregating energy without training, such as Deep Learning methods, for example. The application of the algorithm used public databases, such as REDD (Reference Energy Disaggregation Dataset) and REFIT, in order to implement the solutions presented by the article that published the GSP-based algorithm. The disaggregation was performed, the disaggregated data obtained approximate the results collected individually, but it is necessary to improve the accuracy.