Research On Algorithms In The Recognition Of Arc-Fault Patterns

When a series generates arc fault, it will be harmful to the electric equipment or the lines or the objects beside the electric equipment, and can easily cause the fire accident, at the same time, the current in the circuit is usually not more than maximum allowable current. Therefore, the traditional protection methods of series arc fault can not protect the circuit effectively. Therefore, it is of great significance and practical value to research a kind of algorithms of patterns recognition which can be well used to test the arc fault, and to diagnose the series arc fault in typical load protection.Through a lot of experiments to generate and simulate the series arc fault, the current waveform and data in the circuit of the typical household electrical appliance can be collected in the condition of normal operation, switch plugging, and generating fault arc. By means of wavelet analysis theory, the characters that can describe series arc fault can be extracted to construct the input of BP neural network, that is used to train and test the established network model to recognize fault arc. Finally, it is verified that patterns recognition algorithm of BP neural network has high recognition rate.The current of series fault arc is usually very small, arcing duration is very short, and includes a lot of nonlinear signal, so it is difficult to detect by Fourier transform. With its time-frequency domain localized nature, and adjusting the time and frequency window, mutation information can be extracted from the signal through Wavelet transform. In this Paper, it is proposed the algorithm to detect arc fault based on the wavelet transform and calculation of energy, fault arc signal is decomposed to four components by db1. The character of energy spectrum in fault frequency-band is obtained as the references of distinguishing the aviation arc-fault.