Detecção e classificação de distúrbios em qualidade de energia usando modelos narx neurais

Abstract

The study of the area of power quality (PQ) has grown recently. The increasing use of high power converters and of nonlinear loads cause changes in the electrical signal (current andvoltage), which are often called disturbances, damaging equipment. This work presents two systems, one for detection and another for the classification of power quality (PQ) disturbances, based on NARX neural network models (nonlinear autoregressive with eXogenous inputs). The NARX networks predict the voltage signal value one step ahead and by analyzing the residuals of these models, the disturbance is detected or classified. A total of 6 classes of disturbances were tested: notch, sag, swell, spike harmonic, oscillatory transient, and a class of signals without disturbances to detect the nominal signal. For these signals, tests at different noise levels (40 dB, 50 dB, 60 dB and 70 dB) were carried out. The detector presented fast and unprecedented results in the literature, with the use of 6 to 7 samples of the signal in the detection of disturbances, with an average accuracy of 91.8%, while the classifier achieved with 1/4 of a cycle (64 samples), an average accuracy of 84.2%, showing the effectiveness of the proposed method considering the reduced number of samples used.