Research On Transformer Condition Monitoring System Based On Key Parameters

In recent years, with the rapid development of smart grid all overthe world, research on the intelligent substation is also increasing. TheInternational Electro technical Commission developed IEC61850standard specifically. Under the guidance of this standard, anintelligent component model of transformer, which is the mostimportant primary equipment in the substation, was built in this paper.Software named “Intelligent Transformer Monitoring and DiagnosisSystem” was developed, which includes online monitoring module,online control module and online measurement module. Themonitoring module consists of partial discharge, DGA and windingdeformation. The control module consists of cooling system controland load tap changer control. The measurement module consists oftemperature measurement and load measurement.Partial discharge and DGA were defined as two key parameters ofthe fault diagnosis in power transformers when building thetransformer intelligent component model. Then a transformer faultexperiment based on the key parameters was designed. This paperchose suspended discharge, point discharge, surface discharge and airgap discharge as four research object. These four discharge sources anda simulate transformer were designed at first. The joint experimentincludes two online monitoring techniques, one is partial discharge,and the other is DGA. ERA method and UHF method were used todetect partial discharge in the transformer, and photoacousticspectroscopy online method and gas chromatography offline methodwere used to detect the dissolved gas in the transformer oil. As seen from the experimental results, each of the four dischargepatterns has its own characteristics, but it is difficult to recognize thesefour patterns using experimental results of the separate methods. Inchapter V, a transformer joint fault diagnostic model based on BPartificial neural network. In this neural network, three frequencydomain characteristics of the UHF PD signal including150MHz⁴50MHz,450MHz⁷50MHz,750MHz¹.05GHz and sevenconcentrations of the fault gases:H2, CO, CO₂, C₂H₂, C₂H₄, CH₄andC₂H₆were considered as a10-dimensional input vector. And the codeof four discharge patterns was considered as the4-dimensional outputvector. The node number of the hidden layer in this neural network wasset to30. Twelve sets of experiment data were chose as learningsamples for the BP network. After training, eight different sets ofexperiment data were sent to the BP network, and after calculation theycan be identified accurately.