Research On Arc Discharge Fault Detection Of Oil-immersed Transformer

Oil-immersed transformers are key substation devices that are widely used in power systems.The internal failure of the transformer often causes the transformer to be shut down or even damages the accident,which directly threatens the safety of the grid operation.At present,there are methods for oil failure mass spectrometry detection,partial discharge and vibration detection.However,in the high temperature,high pressure,oil sealed working conditions,due to the harsh working environment,strong noise and other factors,there are still challenges in the existing methods to achieve accurate transformer fault detection and identification.In this paper,the arc discharge fault detection in oil immersed environment is studied for the arc-discharge fault of oil-immersed transformer.The arc light detection,analysis and arc-discharge fault discrimination method based on machine learning method are discussed and verified.The main work and achievements of the thesis include:(1)The spectral composition characteristics of the arc light in the oil immersion environment are determined by experimental methods.The mechanism of arc discharge is analyzed.The arc discharge fault simulation device is built by AC welding machine.The experimental platform of spectrum analysis and the arc light signal acquisition platform are designed and built.Through experiments,it was confirmed that the arc spectrum of the copper material in the oil immersion environment was 700-1000 nm in the infrared band,and 100 fault-free,200 arc-discharge fault sample data were collected.(2)The feature extraction scheme of arc light signal based on wavelet packet decomposition is constructed,and the feature extraction and feature selection of signals are realized.By analyzing the characteristics of signal band energy variation,a 5-layer wavelet packet decomposition is used,and the normalized band energy calculation determines the energy characteristics of the 32 bands of the arc signal.The 32 energy band eigenvalues are further filtered using the relief algorithm.According to the principle of contribution ranking,11 characteristic energy values that can be used for the classification of arc discharge phenomena in oil media are determined.(3)The arc discharge fault detection model of oil-immersed transformer based on support vector machine is constructed,and the model is verified by experiments.Based on the collected samples,six machine learning models including accuracy,sensitivity,specificity,false recognition rate,leak recognition rate and Yoden index were discussed to solve the arc discharge fault detection.Sexuality,and further discussed the impact of model parameters on classification performance.The experimental results show that the support vector machine method has better performance.In the RBF kernel function,the parameter c=0.009765,g=3.4822,The trained model is trained in 300 samples,of which 270 training samples and 30 test samples are averaged by ten-fold crossover,and the detection accuracy,sensitivity,characteristic degree and Yoden index are 96.000% and 97.086%,respectively.95.374% and 92.460%,the misrecognition rate and the leak recognition rate of the detection model are only 4.626% and 2.914%,which proves the validity of the model.The research work of the thesis can be used as a supplement to the fault detection method of the existing oil-immersed transformers.It has certain reference significance for applying the arc light detection method to the internal fault detection of oil-immersed transformers in the future.