Detection of series arc fault based on probabilistic neural network

Abstract

The arc fault includes parallel arc and series arc. The parallel arc fault is characterized by short circuit current and fault current is large, which can be protected by the current circuit breaker. The series arc fault is limited by the load and fault current is small，which cannot be protected by the current circuit breaker. The current waveform of normal operation and arc fault is obtained through experiments, and the characteristic values of wavelet transform are extracted. The characteristic value was input into the probabilistic neural network model. According to UL 1699 standard, arc fault is judged by calculating whether the half-cycle fault number is greater than 8 in 0.5 s. Using MATLAB simulation, 40 groups of test data are selected. 38 groups of test results are correct and 2 groups are wrong. The fault identification rate is 95%, which shows the effectiveness of the method.

Keywords

• arc fault detection
• probalitistic neural network
• feature signal value
• wavelet transform