Design Of Partial Discharge Detection And Identification System For High Voltage Cable

The study of partial discharge(PD)detection can effectively alleviate the instability of power system transmission and distribution.On line detection is an effective method for partial discharge detection of HV cables.However,the main reason for not being a mainstream in industrial applications is that the existing methods for the acquisition and recognition of the signal still need to be improved.In the detection of partial discharge signals,there are many types of interference in the collected partial discharge signals.Some interference signals are very similar to those of the PD signals.It is difficult to remove them from the PD signal.In the PD signal recognition,it is difficult to identify the signal with high similarity.Therefore,this paper focuses on the acquisition and recognition of partial discharge signals from two aspects.The main research in this paper includes the following aspects:(1)Statistical analysis of five different types of 3500 group of sample data are carried out.Explore the relation between peak voltage,discharge of the different types of signals and test voltage.The two-dimensional and three-dimensional phase map of five types of PD signals are analyzed.At the same time,the correlation between the local characteristics is also analyzed.(2)High voltage cable PD System is designed from hardware layer and software layer.In terms of hardware,integrated operational amplifier is selected to amplify the signal based on the characteristics of PD signal and band-pass filter is selelcted to filter out the preliminary.At the same time,oscilloscope is selected for data acquisition and display.The sampling circuit is designed by the IEC 60270 protocol.At the software level,the interface of detection and recognition system is designed by using the MFC interface programming technology.The function of the software includes data collection,data display,data management,data denoising and trend analysis.Meanwhile,data acquisition,visualization and preliminary processing are also completed.(3)Five different types of PD signals are identified by the method of Gradient Boosting Decision Tree(GBDT).By constantly adjusting the parameters of each model,the recognition accuracy is up to 92.5%,while in the same conditions,the recognition accuracy of the SVM and KNN methods are as high as 88.1% and 83.7%,respectively.