Research On Identification Method Of Single-phase Grounding Fault Nature On Transmission Line

The traditional reclosing device will reclose after a predetermined time delay after the fault occurs.When the reclosure occurs in a permanent fault,it will cause a secondary impact to the power system.Compared with traditional reclosing,adaptive reclosing needs to identify the nature of the fault before closing,and only coincides with transient faults to avoid blindness of reclosing.The key of adaptive reclosing is to identify the nature of the fault.Therefore,it is particularly important to study the identification method of single-phase ground fault nature of transmission lines.First,the arc characteristics and the voltage characteristics of the single-phase ground fault of transmission lines are analyzed,and the voltage characteristics of the transient and permanent faults of the transmission line are simulated and analyzed by using ATP.Secondly,a fault feature extraction method based on Local Mean Decomposition(LMD)and Sample Entropy theory is proposed.The extracted fault features are used as the input of BP neural network to identify the fault nature.Realizes the identification of single-phase ground fault nature of transmission lines based on BP neural network.The voltage waveforms of the fault phase of the two faults are different,and the unique convolution kernel of the convolutional neural network can extract deep-level features of the image for image recognition.A convolutional neural network capable of recognizing the nature of single-phase ground faults on transmission lines is built.The fault phase voltage waveform is processed as a grayscale image,which is used as the input of the network.This will reduce the steps of extracting fault features.Finally,a single-phase ground fault nature identification software for transmission lines is designed,which integrates the two fault nature identification methods studied in this paper.It has the functions that users can independently choose the identification method to identify the nature of the fault,update the sample library,and retrain and save the network.The research results show that:using the LMD sample entropy as the fault feature and based on the BP neural network for recognition,the recognition accuracy rate reaches 99%;the recognition accuracy of the recognition method based on convolutional neural network can reach 99.6%,but the required recognition time is closely related to the computer configuration;the software that designed to identify the single-phase ground faults in transmission lines can flexibly apply the two methods in this thesis.