Research Of Online Detection About Insulation Condition Of Capacitive Equipment

The insulation as an important component plays a role in isolating different potential and keeping the normal operation of the electrical equipment, which decides the equipment’s life to a large extent. Usually the number of the capacitive equipment accounts for 40%-50%of the total of the electrical equipment in the transformer substation. So it’s very significant for the safe and reliable operation of power system to ensure good insulation of the electrical equipment. At present, the vast majority of online measuring methods for insulation condition of the electrical equipment choose the dielectric loss factor as the preferred measurement parameter. Considering the dielectric loss value is so small and easily affected by the disturbance, so seeking a measuring method with high precision is still of great significance.In this paper, the calculation method of dielectric loss factor based on Hilbert transform is not affected by the fluctuation of grid frequency and non-integer period sampling, which generates the maximum absolute error in 10e-5 orders of magnitude. Taking into account the harmonic effect on the calculation results, this method can still guarantee a high accuracy and almost the same level of error when the total voltage harmonic distortion is in the normal range. We adopt the method of empirical mode decomposition based on data-predicting extension fulfilled by the nonlinear regression neural network to obtain the fundamental component of the voltage and current signal used to calculate the dielectric loss factor and to process the online monitoring data of the dielectric loss value. This method of signal decomposition is particularly suitable for processing nonlinear, non-stationary data sequence. It’s better than the often used symmetric extremum method and still can accurately decompose the fundamental component under more disturbance and severe distortion of the signal. At the same time in order to reduce the measurement error and to eliminate the impact of noise we adopt a filtering scheme made up of the wavelet denoise and an alternative hybrid filter based on mathematical morphology while preprocessing the sampled data. We employ these methods step by step to process sampled data and to calculate the dielectric loss factor. The simulation results show that the error is in the range as the measurement required and the whole method can achieve the expected accuracy.