Embedded system for electrical load characterization based on artificial neuronal networks in the management of electrical demand in a domotic system

Abstract

A load characterization system was developed in an embedded platform, in order to identify electrical devices used in the home. For the characterization process, the most representative electrical parameters of the different loads were defined, which were used in the training of an artificial neural network implemented in an embedded platform with a network topology with the best performance in terms of computational resources, time of execution and percentage of error. The network topologic had two hidden layers each one with 10 neurons. With the characterization of electrical charges, an intelligent home automation system could be created which can generate savings of up to 23% compared to traditional home automation systems or 69% savings compared to a home without any automation or system control. The proposed demand management system can actively manage the loads due to the knowledge of the elements connected to the network, identifying periods of low consumption which can be related to charging processes completed in mobile phones, laptops or standby mode for televisions. The identification of charges facilitates the implementation of management schemes and control of electric charges.