Study On Feature Parameter Extraction And Pattern Recognition Of Acoustic Signal From Transformer Oil Discharge

Power transformer is one of the most important electric transmission and transformation equipment in modern power system. The insulation discharge is a major factor that causes transformer fault and even induces serious electric power system failure. Transformer oil is an extremely important kind of insulating dielectric in oil-immersed transformer, and the discharge in transformer oil is always accompanied by the release of acoustic energy through the medium in transformer. The acoustic signals reflects the insulation condition of the power transformer to a certain extent. Therefore, it's of great theoretical and practical value to conduct the study of acoustic phenomenon ouccurring in the discharge of transformer oil.The generation mechanism and propagation of the discharge acoustic signal is analyzed based on the phenomenon of acoustic signal in transformer oil discharge. A test platform of discharge in transformer oil is built and four typical discharge models are designed, including needle-needle discharge, needle-plate discharge, creeping discharge and suspended discharge. The wavelet threshold denoising method is adopted to preprocess the discharge acoustic signal. By selecting the appropriate wavelet base function, decomposition levels, threshold selection rules, reset threshold value method and threshold function, ideal result is achieved in the study of discharge acoustic signal preprocessing.The extraction method of multiscale entropy is put forward, and the multiscale entropy parameter can well characterize the complexity of the discharge acoustic signal. The characteristic parameter of actual discharge acoustic signal samples is also constructed, and the multiscale entropy value distribution of different types of transformer oil discharge has its own characteristics. The theory of support vector machine(SVM) is applied to classify the discharge acoustic signal pattern. Through the selection of reasonable classification method and relative parameters, the recognition of the discharge pattern is effectively achieved.