Detection of high impedance faults using neural nets and chaotic degree

From preliminary research, the key characteristic identified for high impedance faults (HIF) was the arc normally associated with these faults. Arcing fault current occurs when voltage is high and increases in proportion to the voltage magnitude; then it decreases as voltage goes down. In this paper, a new HIF detection method that uses a backpropagation neural network as a fault detector is suggested. To analyze arcing fault current properties, the authors divided one cycle fault current into four equal spanned windows according to the voltage phase and applied the fast Fourier transform (FFT) to the current waveform in each window (window length is a quarter of a cycle). FFT magnitudes of the harmonic current were entered as input variables of the backpropagation neural network. In addition, application of the concepts of fractal geometry to analyze the chaotic properties of high impedance fault currents were described and the existence of chaotic behaviors are proved by evaluating the fractal dimension and phase plane. The new detection method is applied to field tested data which are measured in actual 22.9 kV distribution system and shows improved accuracy.