Study On Data Processing Algorithm In Transformer Partial Discharge Monitoring

With the conception of smart grid proposed,researches on smart transform of many basic facilities of power grid like large-capacity power transformers has been a hot point of power grid industry,the stability and insulation performance of power facilities are critical for ensuring their operation safe and sound.Historical data shows that the vast majority of power transformer safety accidents are caused by insulation problems.It is commonly used to judge the insulation performance of the transformer by monitoring a transformer to check if there is partial discharge exist.However,there are always some inevitable undesired signals in the process of data sampling.Therefore,denoising is a big problem to be solved during partial discharge monitoring of transformers.As undesired signals from complex working environment are uncertain,it is difficult to obtain its signal characteristic parameter in advance for targeted filtering.Wavelet is widely used in many fields of industry for signal analysis and shows good time-frequency characteristics.The choice of wavelet basis function and the number of decomposition layers are the key factors for denoising process,which has a great influence on the denoising performance,and modified algorithm that aim at avoiding these problems will be proposed.In this dissertation,primary coverage is as follow:(1)The causes and effects of the partial discharge of the transformer are analyzed firstly,as well the process of discharge,and mathematical model of partial discharge is built via Simulink to help further explaining.Then introduced different applications of data processing algorithm.The origin of undesired signal during detection process is analyzed,comparing the difference of the commonly used denoising methods,modified algorithm is proposed afterward.(2)A soft threshold denoising method based on a complete ensemble empirical mode decomposition with adaptive noise and sample entropy is proposed.By decomposing origin noise-mixed-signal into different modes,and calculating the sample entropy value of each mode to eliminate periodic narrowband signal via setting a threshold.And remaining signal components are mainly useful signal and white noise,furtherly,white noise is eliminated by soft threshold function.The choice of soft threshold in denoising process has a great influence on denoising performance.There may be problem of signal amplitude loss in conventional soft threshold function processing,an modified soft threshold function is proposed here and is proofed to be more useful in signal information retaining and denoising synchronously.(3)Since the empirical mode decomposition algorithm is not so efficient for iterative calculation,modal aliasing and false components,a denoising algorithm based on empirical wavelet is proposed.Similar to previous ideas,using empirical wavelet transformation to extract intrinsic mode functions adaptively,which can save the computing time that spend on iterative calculations.The outcome after subsequent sample entropy processing and soft threshold processing of the modal components shows a better denoising performance on SNR and MSE value of denoised signal,which was verified by comparing with other methods.