Partial Discharge Fault Diagnosis Of Power Transformer Based On Entropy Theory

As the developement of the whele powar indutsry,the voltege level of power equipmant is constantly improveing,and people have pute forwerd more and more stringment requirements for the operation safety and reliability of power equipment.Partiel discharge detection is an nendestructive,traumatic test type that is increasingly widely used.This paper is studied from two perspectives of signal denoising and pattern recognition.The reasan for the partial dischrage of the trensformer is that the elactric fiald strength of an internal medium is too high to reach the critical breakdown electric field of the medium.It destroys the molecular structure in the insulation layer and then destroys the insulation layer.Differeant discharege faults have differeant effeacts on the equipement,so the identaification of partial discharage types can previde valuable technical suppaort for the maintenantce of the transforamer.The denaising method with complate antegrated empirical made decomposition（CEEMDAN）and appraximate entropy（ApEn）is used to analyze the partial discharge of transformer from both simulation and field.First the noise-containing signal is CEEMDAN decomposition,then find approximate entropy for each component to eliminate the noise component.Then the correlation coefficient is calculated for the components under retention,the components with small correlation are removed and the remaining components are reconstructed.The denaising of simalated signals with differeant signal to naise ratios is more stacle than conventional methads.The field signal is denoised,and tha resultas show that the discherge pulse cain be effectaively extractced.This paper combines empirical wavelet transform（EWT）and multiscale quantum entropy（MQE）.First,the EWT decomposition of the experimental signal is performed,and the decomposition results can contain rich partial discharge information.MQE is then selected as the feature parameter,and MQE can accurately characterize different partial discharge types.Finally,local tangent space arrangement（LTSA）and hierarchical clustering（HAC）are used for pattern recognition.Validated with different types of PD experimental signals and compared with traditional K-means and Descan,the results demonstrated excellent diagnostic performance.