A fault line selection algorithm using neural network based on S-transform energy

An approach to detect fault line in distribution network using neural network based on S-transform energy is proposed und after analyzing the variance of fault characteristic frequency of zero sequence current in each feeder line of overhead line and underground cable mixed lines. In order to avoid the effect of TA's disconnection angle, the short window data of first 1/4 cycle are selected. The S-transform is carried out to determine the main characteristic frequency of fault zero sequence current, and taking the Short Window energy of the main characteristic frequency as the target input to form BP neural network model, thus the fault line can be detected adaptively. State component and various noises can be filtered out utilizing S-transform to determine the main characteristic frequency. Fault detecting margin can be enhanced by adjusting the weight of criterion through neural network training accurately. The theoretic analysis and simulations demonstrate the feasibility and validity of this approach, also the problem that training time is too long and network result is too complex is well solved when using traditional neural network to detect fault line.