Research On Fast Fault Diagnosis Of Transformer Based On Fluorescence Spectrum

Power transformer is one of the most important equipment in the power industry.Its stability and reliability will affect the whole power system.Nowadays,most transformers are oil immersed,which is expensive and prone to failure.Therefore,it is necessary to conduct regular maintenance on the transformer to help troubleshoot.At present,the commonly used transformer fault diagnosis method is mainly the detection of transformer oil,but the traditional detection method of transformer oil will inevitably produce a long test cycle,and manual operation will also lead to errors.To solve this problem,this paper focuses on the fast and accurate diagnosis of transformer,and proposes a fast diagnosis model of transformer fault based on fluorescence spectrum,which uses the laser induced fluorescence(LIF)technology with short detection time(about 2-5 s)combined with the whale optimization algorithm(WOA)extreme learning machine(ELM)algorithm with fast detection speed and high recognition accuracy,It helps to realize the fast and effective diagnosis of transformer fault.In this paper,the common fault types of power transformer are selected,and the oil samples are divided into four cases: normal transformer oil sample,lightning strike fault,water inflow damage fault and short circuit fault.In the laboratory,200 groups of fluorescence spectral data were collected for each transformer oil sample,and a total of800 groups of spectral data were obtained.In the research process,because the data dimension reaches 2048 and the redundancy is high,the original data is preprocessed first(including denoising and dimension reduction).Through the comparison of signal-tonoise ratio,the S-G filtering algorithm with better denoising performance is used as the experimental filter.In the aspect of dimension reduction preprocessing,the principal component analysis(PCA)method of maximum expectation optimization is selected for feature extraction.The combination of maximum expectation algorithm and PCA algorithm can effectively extract feature data information,improve the accuracy of classification and help improve the detection speed.The fast running ELM algorithm is selected as the classification algorithm.At the same time,in order to better improve the detection accuracy,swarm intelligence optimization is carried out on the original ELM model.Finally,WOA-ELM algorithm with high recognition accuracy and fast detection speed is selected as the recognition algorithm of this research model.The average training set accuracy can reach 99.67%,the average test set accuracy is 99.50%,and the average classification detection time is only 1.525 s,It can ensure the accurate classification of transformer faults while reducing the time cost.In this paper,a fast and accurate transformer fault identification system based on fluorescence spectrum is proposed,which can not only judge the fault category in time when the transformer is regularly overhauled,but also provide a research foundation for the subsequent creation of a more complete power transformer fluorescence spectrum database and a complete transformer fault diagnosis device,helping power enterprises save economic costs and effectively ensure the smooth operation of the power system.Figure[41] Table[8] Reference[81]...