Research On DC Arc Fault Detection Technology Based On BP Neural Network

In the application of photovoltaic system,aerospace,electric vehicles,large room and other large power DC electrical equipment,DC arc fault caused by equipment loss,insulation damage or loose joints will leads to unimaginable consequences such as fire.Because the characteristics of DC and AC arc fault are very different,it is important to study the effective method of DC arc fault detection for the safe use of DC equipment.This thesis researches on the DC arc fault detection technology,presents a DC arc fault detection method based on BP neural network,which can judge DC arc fault according to the input current characteristics.The thesis uses the wavelet transform to reduce the noise of the original current sampling data,analyzes DC arc fault characteristics of the current data in time domain,frequency domain and time-frequency domain respectively with FFT and wavelet,and determines some current features which can be used as inputs in BP neural network to detect DC arc fault.The thesis designs a BP neural network model for DC arc fault detection,determines its parameters such as the numbers of input neurons,hidden layer neuron and output layer neuron,and optimizes its initial weights with genetic algorithm aiming at the problem of slow convergence during the training of BP neural network,finally proposes a complete BP neural network training procedure using genetic algorithm to optimize its initial weights.The BP neural network is trained with the DC arc fault features sample,after train it is used to detect the DC arc fault.Experimental results show that the method proposed in this thesis can achieve the expected requirements of the detection accuracy and the false detection rate of DC arc fault.