Identificação de equipamentos elétricos residenciais

Abstract

The electricity consumption in homes has been increasing continuously along of the decades and actions to reduce these consumption should be taken. An interesting solution is provide to the consumers, and also to the energy company, the feedback about the consumption of each electrical appliance. Therefore, this work consists in the development of a system, based on Neural Networks, able to identify the equipment through of yours electrical characteristics like active and reactive power. The topology used in the neural network is Multi-Layer Perceptron structured in three layers as the algorithm training uses the backpropagation method. Among the steps of the identification system, preprocessing proved to be crucial to the learning process of the network, in the same way, the selection of variables reduced the complexity of the network, reducing the predictive attributesand improving the RNA processing. The analysis of the signatures used was steady-state, based on a dataset with samples recorded every minute, considering the period of 24 hours for each signature. From this approach all loads were recognized, getting an accuracy above 93% for the overall system and a rating precision of approximately 88% (sic).