Research On Small Current Grounding Fault Detection Based On Wavelet Transform And Neural Networks

Small current grounding fault detection is one of important technology to ensure the power system running with the stability and reliability. It is urgent to be solved in order to meet the requirements of nation safety, industry automation, agriculture and modern life.This dissertation focuses on the study about how to detect the Small current grounding fault and how to decide the fault line in neutral power system. Based on the analysis about characteristics of the grounding fault and detection method that had been presented before, the wavelet transform and neural network was used to find the more correct method and more effective measures to detect the grounding fault. In the last, an integrated detection method was presented. The main contents are as follows:According to the actual structure of power system and based on electromagnetic simulation software EMTP-ATP, the simulation model of metallic grounding and especially for arc grounding fault in small current system was set up. The structures and parameters can be configured according to the actual needs of the simulation. The establishment of models was very important for the follow-up in-depth study of small current grounding fault detection method. It can provide all the effective data and support.Study how to use of wavelet packet decomposition method to extract the features of faulty signals, and how to constitute the feature vectors. Then proposed an integrated detection method based on data fusion. First, give the method use of wavelet packet decomposition to extract the steady-state, transient and harmonic features, and to verify its reasonableness and adequacy. Second, an integrated detection method to detect the grounding fault was detailed described and the above method was validated by the simulation.A novel method to detect and decide grounding fault line based on energy changes was presented. The zero sequence current signal of each line was decomposed by use of wavelet transform multi-resolution analysis. Calculate the energy with the coefficients of each band that contain stationary fundamental and harmonic signals, compare the energy change of each line before grounding and after, the maximum change line was identified as fault line. The simulation results have been verified that the energy change method presented here was most effectively.A method based on neural network to measure the harmonics was presented. By the way of WPT, a method based on neural netwok to detect the harmonics was studied. The original measure data of power system was transform into some sub-band, the coefficients of each sub-band form one sub-vector and as the input of neural network, the contents of harmonics was given by the output of neural networks. Simulation results has showed that the present harmonic measuring method was very accuracy and useable. Then based on the method above, a synthetic method to detect the small current grounding fault based on neural network was presented. The method about how to get the feature vectors and how to train and test the neural network was described. The steady-state, transient characteristic, harmonic feature and the energy change have been integrated. The neural networks was trained and simulated, in different condition, different parameter and position. So that using the feature vector as the input of neural and the fault line can be given in the output. Simulation result has proved that kind of detection method was more accuracy, reliability and suitability.