Research On On-line Partial Discharge Diagnosis Technology Of Cable Terminal In Distribution Network

In distribution network,cable terminal insulation fault is one of the main causes of safety accidents,and partial discharge(PD)is closely related to insulation fault.Partial discharge on-line monitoring device can detect and identify partial discharge signals,which can be used as an important basis for evaluating the insulation condition of cable terminals.However,the existing monitoring equipment usually uses expensive high-speed data acquisition card for data sampling,which makes the cost of the monitoring system very high and difficult to be widely used.In order to solve the above problems,an on-line partial discharge monitoring device based on discrete level signal acquisition is designed in this thesis.The device uses DSP as the signal processor,collects the partial discharge signal based on the high frequency current transformer(HFCT),converts the continuous PD signal with frequency up to tens of megahertz into discrete level signal and transmits it to FPGA through the signal processing module,and finally transmits the data to DSP through serial communication for partial discharge type identification.At the same time,in order to avoid the electromagnetic interference between strong current and weak current in the ring main unit of distribution network,a wireless energy acquisition device is designed to provide power for the on-line monitoring device by collecting the microwave energy in the environment.Through this method,reliable electrical isolation is realized and stable power supply is played.In order to realize the effective and accurate identification of partial discharge types of cable terminals in distribution network,four typical partial discharge models of cable terminals are designed.Through a large number of discharge tests,the PRPD characteristic spectra and Δt characteristic spectrum under different discharge models are constructed.By analyzing and comparing the spectra of different discharge models,the PRPD characteristic spectrum and Δt characteristic spectrum are further grayed,and the fractal features,moment features and texture features of the image are extracted.Due to the correlation and information redundancy between the feature parameters,the dimension of feature parameters is reduced and optimized based on PCA method,and new feature parameters are obtained.Finally,the type of PD signal is recognized based on multi-kernel multi-class relevance vector machine(MMRVM),and the particle swarm optimization(PSO)is selected to optimize the core parameters.The difference of PD signal recognition between it and nearest neighbor propagation algorithm(AP),clustering algorithm based on Euclidean distance(K-means)and back propagation neural network(BPNN)is compared and analyzed.The results show that the total recognition rates of MMRVM classifier according to PRPD spectrum and Δt spectrum are 98.22% and 97.92% respectively,which are higher than the other three classifiers,and the recognition effect of MMRVM classifier is significantly better than the other three classifiers for both aggregated data and scattered data,which has a certain application value.