Fault Location For Underground Distribution Network Based On Wavelet Neural Network

On account of the difficult problem that feeder fault location in undergrounddistribution network is not easy to solve, this thesis put forwards an accurate faultlocation method which utilizes the fault transient state information extracted by waveletpacket analysis based on the simulation analysis of characteristics of the steady andtransient state waveforms collected from single-phase ground fault that happens in cableline of the underground distribution network. Considering the mapping relationship ofthe wavelet packet decomposition modulus maximum of specific frequency bands andthe fault point distance, the cubic b-transect wavelet is used to realize binary waveletpacket decomposition so as to extract modulus maxima of specific frequency bands, andthe strong nonlinear fitting ability and generalization ability of the neural networks areexploited to fit the above mapping relationship to achieve the purpose of fault location.On the basis of analysis of the inadequacy of “loose” type wavelet neural networkapplied in the fault location, this thesis constructs the “tight” type wavelet neuralnetwork and the improved BP algorithm wavelet neural network is also proposed. Thegenetic algorithm and the particle swarm optimization algorithm are used respectivelyto optimize wavelet neural network and RBF neural network parameters, and theoptimized networks are used for fault location. The simulation results confirm thatcompared to the traditional “loose” type wavelet neural network,“tight” type waveletneural network based on genetic algorithm and the RBF wavelet neural network basedon particle swarm optimization algorithm can realize the accuracy, reliability andstability of fault location, the ranging accuracy is guaranteed also.