Engineering Application Of Oil Chromatographic Analysis In Power Transformer Fault Diagnosis

Power transformer is one of the most important electrical equipments in the electric system which bears the responsibility of voltage converting,power transmission and distribution in power system.So its safe operating state attaches importance to power system's reliability,high quality and economic operation.The times and frequency of transformer fault must be reduced as far as possible.Transformer fault diagnosis based on oil chromatogram analysis is the key to reduce damage rate itself and enhance the efficiency of applying power,which is helpful to detect the potential transformer faults and carry out condition-based maintenance in time.Although the condition of transformer internal fault can be reflected effectively with dissolved gas in transformer oil,the accuracy of the results may be affected by the innate reasons of judgment method using fault diagnosis based on oil chromatogram analysis.In order to improve the accuracy of fault diagnosis,studies have been done in the following several aspects in this paper:(1)Because of the incomplete code,diagnosis of transformer internal fault tends to be left out and misjudged when using traditional ratio method.On the basis of the original ratio method and combined with Gene Expression Programming(GEP),A kind of fault diagnosis method diagnosing step by step is proposed.With that method the type of fault,overheat fault or discharging fault,is distinguished roughly first,then the fault point temperature or the discharging intensity is judged for specific fault type.A lot of Examples analysis proves that,the omission of judge is reduced effectively and the accuracy is improved when applying this method.(2)A 5-11-5 BP neural network model of transformer fault diagnosis is established based on the characteristics of transformer fault gases and fault types.According to the poor convergence of network due to discrete inputting datum,a normalization method based on cumulative frequency is proposed.Under the same weights and threshold and other parameters,the comparative studies have been carried out on the transformer fault diagnosis with cumulative frequency and other methods.A conclusion can be drawn from the comparative research that the classification performance of BP network can be improved effectively applying cumulative frequency normalization method compared withthe other three methods,which has very important reference values on the study of neural network's application.(3)Software of transformer fault diagnosis is developed by applying JAVA programming language based on the two methods above,transformer faults that may happen can be estimated by entering the component concentrations of dissolved gases in oil using that software,which brings greater convenience to engineers.And the effectiveness of the software is verified by the analysis of a typical faulty transformer.