Detection And Analysis Of Insulation Oil Based On THz-TDS In Power Transformer

Power transformer is a key equipment for power grid safety.In order to ensure the safety of power supply,it is very essential to diagonize and detect the fault category of power transformer in time.Currently,analysis on transformer oil,like gas dissolved in oil,are applied on the fault diagnosis of power.However,those technologies are mainly based on gas chromatography,which is an infrared spectroscopy detection of dissolved gas in transformer oil.The traditional analysis methods are complex operations or lower resolution to discover the transformer faults.Therefore,it is dramatically important to find out a new analysis method,which is more efficient and effective.In the past decades,terahertz-time domain spectrometer(THz-TDS)is gradually developed.It is a challenging attempt to analyze transformer faults by using this technique.In this paper,wide-band terahertz time-domain spectroscopy and stoichiometry are used to analyze transformer insulation oil.The main research contents of this paper include:(1)The noise problem of terahertz time-domain spectral signals.In order to reduce the noise of terahertz time-domain spectral signals,this paper has carried out analysis from the aspects of hardware and software.In terms of hardware,this paper analyzes the influence of time constant of phase-locked amplifier on noise.In terms of software,this paper studies two noise reduction filtering methods: one is bilateral filtering algorithm,and the other is wavelet transform algorithm.Four conventional denoising algorithms have been studied and analyzed,such as fixed threshold method,unbiased estimation method and minimum maximum variance method and heuristic threshold method,the comparative analysis of four kinds of threshold algorithm for denoising effects.Because of this,a new method was put forward based on soft and hard threshold functions.By combining the advantages of both,the noise reduction effect is improved obviously.(2)Identifing the chemical structure and physical characteristics of substances in transformer insulation oil through optical parameter analysis.This paper first introduced several methods of extracting optical parameters by terahertz transmission time domain spectrum,including analytic method based on Fresnel formula,total variation minimization method,collinear space method and two-layer structure extraction method.Then several common methods of extracting optical parameters by terahertz reflection time domain spectrum are introduced,including: kramer-kraniger analysis method,Fresnel formula method and attenuated total reflection method.Finally,the characteristics of various methods are analyzed,and the optical parameters extraction and calculation methods of terahertz spectrum suitable for transformer insulation oil analysis are determined.According to the characteristics of transformer insulation oil,a terahertz optical parameter extraction method based on TD optical model is proposed.By using this method,the insulation oil of transformer with 5 different working years was tested.The experimental results show that with the prolongation of working time,lower molecular hydrocarbons and impurities will be decomposed in transformer insulation oil,which provides a good condition for judging transformer faults through the analysis of transformer insulation oil.(3)By using terahertz time-domain spectroscopy and stoichiometry,the insulation oil of three typical transformer faults,namely thermal fault,electrical fault and local damp fault,were qualitatively analyzed.In order to analyze the transformer fault more carefully,the thermal fault is simulated and three kinds of oil samples with different thermalization temperature are prepared.In this paper,three common transformer fault models are designed and built,and the simulation results are sampled and analyzed.Three kinds of algorithms,like chaotic particle swarm optimization algorithm,genetic algorithm and parameters of standard particle swarm optimization algorithm of support vector machine(SVM)models were proposed on the analysis of three different thermalization temperature and four kinds of discharge models.According on the analysis of these SVM algorithms,a particle swarm optimization support vector machine based on annealing parameters was put forward.It has been applied on the analysis of thermalization and electrical faults,which the accuracy reached 98.55% and 98.18%,respectively.(4)Terahertz time-domain spectroscopy was used to detect hydrocarbons and other substances produced during the aging of transformer insulating oil.Due to the high transmittance of terahertz wave and the relatively sensitive characteristics of intermolecular vibration and rotation,the changes of functional groups during the deterioration of insulating oil could be detected and observed.In order to further study the prediction of moisture content of transformer insulating oil in partial damp failure,a simulation test was set up in the laboratory,and terahertz absorption spectrum was analyzed by partial least square method to realize the quantitative prediction of moisture content of transformer insulating oil based on terahertz time-domain spectroscopy.(5)In order to take advantage of terahertz time-domain spectroscopy on identification of transformer thermal fault,electrical fault and local damped fault,partial least squares discriminant analysis method has been introduced in this paper.Based on the analysis of partial least squares discriminant method,this paper introduces the conventional interval partial least square method(I-PLS),moving winow partial least square method(MW-PLS)and backward interval partial least square method(BIPLS).The fuzzy clustering preferred variable partial least square method(FCM-PLADA)is proposed.In order to compare the effect of the classification of the four methods of transformer faults,this paper collected three simulated fault of transformer insulation oil samples of terahertz signal.By extracting the characteristics of the absorption spectra and refractive index of samples,the absorption spectra are applied for the verification for the proposed algorithm which is to identify three different fault types.The experimental found the classification accuracy of the FCM-PLS-DA model proposed in this chapter is significantly better than the other three pls-da methods,obtaining the ideal effect of 100% classification accuracy.It provides a new method for transformer fault determination.The conventional method,gas chromatography,is compared with the results in this paper.The results show that the method based on terahertz time-domain spectrum analysis of transformer insulation oil studied in this project is the same as the existing methods in identifying typical transformer faults and has a good application prospect.The present dissertation includes 58 figures,26 tables and 160 references.