The Research On Grounding Grids Fault Diagnosis Based On Neural Network

Grounding grids is an indispensable establishment in power stations andsubstations. It plays a vital role in keeping stablility and safety of power system, andin ensuring the security of operators and power equipments,and it can also offerreference ground potential of power system. However, as the grounding grids isburied in the earth, the electric performance parameters will become worse for somespecial factors such as harsh conditions, and inevitable corrosion,which will causethe ground connect broken, and directly threat the safety of operati on of the powersystem. So the research for locating the grounding grids fault rapidly has veryimportant significance.Based on the electric-network analysis method in the diagnosis of groundinggrids, this paper constructs grounding grids as pure resistance model in the DCresource. Node voltages are obtained when faults occur in the grids. By defining thenormalized voltages as input samples to train the network, so the trained network canbe used in fault diagnosis.The fault feature of the pure resistance model is not enough to establish stabilityequation, for accurate diagnosis results. Therefore, the grid conductors are tainded asπ model which takes the grid conductor resistance and reactance into account and thenode voltage waveform can be gotten including its amplitude and frequency.Furthermore, wavelet packet decomposition and reconstruction is used to dispose thewaveform signals and construct input samples of the neural network, and the accuratediagnosis results can be obtained by the trained neural network.To solve the problems of the neural network·s slow convergence speed and localtraps, ant colony algorithm is used to optimize the network, then the network is usedto diagnose the fault of grounding grids. The neural network has faster convergencespeed and higher correct recognition rate after optimized.