Research On On-line Monitoring Data Analysis And Insulation Diagnosis Of Dielectric Loss Factor

In order to realize insulation state of electric power equipments and avoid or reduce electric equipments breakage and non-plan power cut due to insulation fault, it is essential to carry out insulation detect and diagnosis. In this dissertation, approaches for the high accuracy measurement of tanδ, trend extracting of the on-line data, analysis of the data regularity and diagnosis equipments are investigated, the main achievements are as follows:1. As sine wave parameter method could not eliminate the on-line monitor data pulse disturbance effectively and hence leads to dielectric loss angle inaccurate, a novel time-space filtering method based on empirical mode decomposition (EMD) for the measurement of dielectric loss angle is applied. Use EMD to decompose sample signals under strong noise background; and according to the spectrum characteristics of obtained intrinsic mode components the selective filtering is conducted, then the fundamental component is extracted and to calculate the dielectric loss angle combining with sine wave parameter method.2. Dielectric loss factor on-line monitoring data fluctuation scope is quite wide, so it is very difficult to make out the change trend. This paper advances the trend extracting method based on empirical mode decomposition. EMD approach can separate signals which contain frequencies from high to low and direct current component efficiently by thrice strip function. By comparing the results with wavelet transform and morphological filter. Mathematical basic expression of tanδis established, and it supplies ample condition with insulation diagnosis.3. For equipment insulation characteristic is influenced by many environment factors, impact extent of correspond factors changes as the insulation status changes, wavelet entropy is employed to diagnose insulation fault. Wavelet entropy combines the time-frequency localization ability of wavelet analysis and the system state ability of entropy to token. Wavelet energy entropy is calculated as system characteristic parameter and then distinguishes from trend variety of different signals. Using the dominant variable to fault diagnosis which is the adjacent quantity to tanδ.