A Line-selection Method Of Data Fusion Based On Neural Network

The small current grounding method mainly adopt the way of isolated neutral, neutral grounding by the arc extinction and neutral grounding with resistance in our country of low voltage distribution network. This method avoids power interruption caused by protection device tripped while the small current grounding fault occurs (also know as single phase to ground fault). And then could increasing the stability and continuity of the power supply. But because the small steady-state component of the small current single phase to ground fault, the weak current fault and the lines of different lengths and complex structural and so on, make ground fault location more difficult.According to the statistics of department concerned, the highest probability fault is single phase ground fault in power distribution network, about80%of the up power system failure. It is the scientific workers that need to solve the difficult problems about how to determine the fault line safely, quickly and accurately, which is often encountered in practice. The current of the small current system fault is small, vulnerable to the interference of the impacts of the changing parameters of the line itself and the natural environmental changes. Over the years there are many research about a variety of the fault of the small current grounding characteristic, and put forward a number of line selection principle and the way of implementation. But each method has its scope of applications, not widely utilized. In recent years, with the rapid development of computer technology, electronics technology and information technology, making transient fault signals can be quickly extracted, and providing the precondition for the fault line selection.In order to overcome the shortcomings such as bad effect of the single line selection, limited range of application and so on. In this paper, on the basis of the analysis of the fault characteristics of the small current single phase ground system, using the fifth harmonic of the zero-sequence current, the line active power and the wavelet energy of the zero-sequence current as characteristic quantify of the fault line selection, and using of BP neural network multi-criterion fusion line selection. Through the MATLAB simulation test, we got an ideal result of the location, and solved the problem of the low accuracy of a single location in a better way.