Research On Aging Diagnosis Of Insulating Oil Paper Based On Terahertz Time-domain Spectroscop

The Transform is the core part of the power system,which will seriously affect the stability of the power grid if it fails while working.The oil-paper insulation system is the key to the stable operation of the transformer.How to accurately diagnose the aging degree of oil-paper insulation materials is one of the hotspots of current power system research.Timely grasp the aging status of the transformer insulation is of great significance to ensure the safe operation of equipment and power grid.At present,gas chromatography is the preferred method to detect the aging of oil-paper insulation,which can accurately identify the insulation aging of the transformer.However,the operation and analysis of this method is too complex,which requires a lot of time and manpower.Terahertz timedomain spectroscopy is a new detection technology developed in recent years,with high signal-to-noise ratio,low energy,and rich physical and chemical information,which can perform high-precision analysis and rapid safety detection of transformer insulating oil.Therefore,on the basis of terahertz time-domain spectroscopy measurement,combined with the fast recognition characteristics of Support Vector Machines algorithm and Neural Network algorithm,this dissertation has carried out a classification study on the aging stage of transformer insulating oil paper.The main work and achievements are as follows:(1)By simulating the aging process of insulating paper under the influence of different factors,the variation of polymerization degree and terahertz optical parameters of insulating paper in the aging process was explored.By changing temperature,moisture,antioxidant and other factors,samples of different aging types and stages were prepared to simulate the changes of actual oil-paper insulation In reality.The dynamic mechanism in the aging process was further explored,and the aging stage of the insulating oil paper in the aging process was determined.(2)According to the degree of polymerization of insulating paper,the aging samples were divided into three stages,and 108 experimental samples of different aging stages were prepared and tested by terahertz spectroscopy.Taking the terahertz absorption coefficient of the oil sample as the input,PSO-SVM,FA-SVM,and FAPSO-SVM classification and diagnosis models are constructed by introducing Particle Swarm Optimization and Firefly Algorithm to optimize parameters of SVM.As a result,the PSO-SVM model can well identify the three aging stages of insulating oil.Through analysis,the FAPSO-SVM model has the best classification accuracy.The model can well identify the three aging stages of insulating oil,with classification accuracy of 91.4%for its training set and 92.6%for its test set.(3)The BP Neural Network diagnostic model and the RBF Neural Network diagnostic model are constructed,and the similarities and differences between the two neural network algorithms are analyzed.After diagnosing the experimental oil samples,the classification accuracy of the training and prediction sets of the BP Neural Network diagnostic model is 93.8%and 92.6%,and the classification accuracy of the training and prediction sets of the RBF neural network diagnostic model is 97.5%and 92.3%.The classification accuracy of five classification models were compared.After comprehensive analysis,the classification accuracy of FA-SVM model,FAPSO-SVM model,and RBF Neural Network were superior to the other two types,with a classification accuracy of 100%in the middle and late stages of Insulation aging.