Rough set neural network based fault line detection for neutral non-effectively grounded system

A novel fault line detection method based on rough set neural network is proposed to avoid the longtime training and the complicated model structure for the neural network based fault line detection model. Through performing single-phase-to-earth fault experiments by means of the ATP-EMTP simulation program, the zero sequence currents of every line are obtained. The fault features are extracted from zero sequence currents by using wavelet transform method, the fifth harmonic current method, zero sequence current active component method and fundamental current component amplitude comparison. These fault features are transformed into fault measures according to properties of each fault feature. The fault measures of lines make up of the original data table and these discretized data construct decision table. Then the decision table is reduced through attribute reduction and rule reduction. The reduced condition attributes are regarded as inputs of the neural network and the reduced samples are used to training neural network. The neural network model trained can realize fault line detection. The simulation results show that the method reaches higher training speed and lower error rate.