Aplicação de Rede Bayesiana para detecção não intrusiva de acionamento de cargas elétricas

Abstract

Monitoring of electric charges aims to identify the activation of electric equipment based on data about its consumption. This process is performed by data acquisition systems, which may be intrusive or non-intrusive and applied at any stage of power generation and distribution systems. The implementation of a non-intrusive monitoring system is possible whenever a home energy meter collects real-time electrical charge data. The collected data feed algorithms, which, after being trained, are able to recognize and identify electrical charges. The system developed in this work is able to deal with uncertainty in the identification of loads applying artificial intelligence techniques known as Bayesian networks. They are oriented acyclic graphs that express, through tables of conditional probability, the degree of belief in certain facts. The exact inference in such networks, however simple, is almost always intractable. Thus, in this work we apply approximate inference by Monte Carlo Markov chain simulation. A complete system is developed and applied to a simplified network. The results show that the methodology employed is adequate for the detection of load activation.