The Research Of Single-Phase-To-Ground Fault Location In A Distribution Network Based On Wavelet Neural Network

According to the electricity sector statistics, the probability of single-phase ground fault usually accounted for more than60%of the total faults in distribution network system, even as high as80%. It has become problem that brook no delay in distribution network system, so how to find out the fault line and find out the accurate fault section, which is to isolate and cut off the fault immediately. It is also become a difficult problem and needed to be solved urgently for scientific research workers. This will be a very important practical value to the country’s whole power system.Neutral non-effective grounding mode by the arc extinction (small current neutral grounding mode) is the most common grounding type of low and medium voltage distribution network in our country. When the single-phase ground fault occurs in this mode, due to the weak ground current and the numerous line branch of the distribution network. They have not obtain a very good solution that line selection and location in the single-phase grounding fault system, this paper make the following relevant work on the basis of research about fault line selection and fault location for power distribution that have been done at home and abroad.1. Part about fault line selection:this paper select the wavelet energy, the fifth harmonic and the active power energy as the input parameters of the fault selection, and define the fault measurement function of them, based on analysis the steady and transient characteristics of the small current neutral grounding system. At last, it use BP neural network on basis of the three characteristics to realize fault line selection. And using MATLAB software simulate a large number of simulation data, it prove that the method has fairly good alignment results. This fault line selection method has resolved the problem, which is the previous single line selection method is susceptible to interference, the accuracy is not high and its small application scope.2. Part about fault location:after completing the line selection, the following paper analysis wavelet packet decomposition theory and BP neural network first, then it constructs a compact wavelet neural network, in which wavelet function of the excitation function in the hidden layer has instead of the traditional BP neural network the excitation function of the hidden layer. And it use the wavelet energy, active power energy and reactive power energy of the transient zero sequence current as the characteristics of data fusion for fault location. And it adopt a momentum item and a new weight initialization method in its training process, which is to improve the network learning efficiency of the network. At last, it also use MATLAB software proves the correctness and reliability of this method.