Wavelet Neural Network With Feeder Single-phase Ground Fault Location Research

With the development of power system, more and more security, reliability and economy are demanded in current supply. Precise fault location technology is becoming important. It is crucial to select the fault line and set out the distance of the fault point. It is not only economizing manual effort but also assuring rapid supply recovery, so it does great importance to economic life and politics.Distribution power systems in our country are almost ineffective grounded systems, it is called Small Current Neutral System(SCNS). But the unique characteristics of ground-faults of the SCNS make it difficult to locate exactly the fault point of transmission lines. With the requirement of improving the reliability of power supply and the fast development of the distributed power system automation, the study of fault location becomes more and more important.This paper analyses the steady and transient character of line fault when the neutral point connects to ground in different ways. From the angle of signal processing, the wavelet analysis theory and the artificial neural networks are introduced to fault location. The new concept is presented, which is to change the reference signal for analyzing and processing from the steady signal to the transient signal in a fault case. A new criterion for earth fault line selection based on wavelet packets is presented: Based on the principle of maximum energy it selects the special frequency band that has the major characteristics of transient capacitance current. Comparing the polarity of the special frequency band of each line with the polarity of the zero sequence voltage's first half wave, it can detect the fault line. The method using wavelet singularity detection theory to extract the fault time is discussed. The wavelet neural network(WNN) model, being advantageous to the adaptive features extraction from the fault transient signal sampling, is constructed in term of the mathematical modeling. This model uses wavelets functions as the NN's basis function, and is provided with the advantage of the adaptive extraction of the specific time-frequency window feature components mapping the fault distance. The phase-to-ground fault distance method of feeder lines in the distributed power system based on the WNN is presented. By means of the simulation of the typical 10kv system, we verify the validity of the algorithms.