Study On Degradation Assessment Of Oil-Paper Insulation And Feature Extracting Of Partial Discharge Of Transformer Based On Chaos Theory

Partial Discharge (PD) has many influences on the oil-paper insulation including, first of all, the destruction of the media by generated high-energy particles which can cut off the molecular chain, secondly, the temperature rise of local part and the sudden increase of the dielectric loss, and finally, the oxidation and corrosion of medium surface by generated active gases. As a result the process of vicious circle between the destruction of insulation and PD will eventually lead to the damage of insulation. Meanwhile, PD is one of the most direct reference features in the monitoring and evaluation of insulation condition, and the practical experience shows that the degree of defect and deterioration of insulation material has a tight connect with the characteristic of PD. Therefore, the monitoring and feature extraction of PD signals has contribution to find out the latent defects and identify the degree of the degradation of local defects in oil-paper insulation, which would protect the operational stability and security of electrical equipment.An oil-paper insulation model which can simulate the insulation among the windings of transformer is designed in this paper. At the same time, the multi-sample damage experimental platform which can conduct damage experiments to a group of samples under the room temperature concurrently is constructed in this paper. Through analysis of the damage time at different voltages utilizing the two-parameter Weibull model, the parameter of lifetime model of experimental samples is obtained. Moreover, the needed degradation voltage of PD is estimated on the basis of lifetime model. At the same time, the method of constant voltage is used to collect the ultra-high-frequency (UHF) PD signal during the process from the beginning of PD to the damage of oil-paper insulation. The characteristic parameters of partial discharge in the degradation process of oil-paper insulation are extracted by applying chaos theory and the changeable discipline of these parameters is also analyzed. Finally, BP artificial neural network of which the extracted chaotic characteristic parameters are used as input is introduced to indentify the degradation degree of the defects.The main conclusions of the paper are as follows:①Results of damage experiments on the winding model of oil-paper insulation at different voltages show that the damage time have a good fitting with the distribution of the two-parameter Weibull model. Therefore, the two-parameter Weibull model can be used to process the damage data so that the lifetime model of inverse power law of oil-paper insulation L=8.8×1015U-12.94 can be obatained.②Chaotic characteristic analysis is carried out to the time series of UHF PD electricity signals, which constructs strange attractors of time series and extracts three chaotic characteristic parameters such as the largest Lyapunov exponent of attractors, correlation fractal dimension and Kolmogorov entropy. Then, the analysis results confirm the existence of chaos phenomenon in the process of UHF PD of oil-paper insulation.③Chaotic characteristic parameters of PD time series are used as the characteristic parameters of assessment of partial degradation of oil-paper insulation. Results of analysis to the time series in the process of degradation of the oil-paper insulation under PD condition demonstrate that the largest Lyapunov exponent has an increasing trend but certain fluctuate with the development of degradation of local defect, the correlation dimension has no monotonous trend but very complicated nonlinear change, and Kolmogorov entropy has a rapid decreasing at the beginning and increasing at later period.④By comparative analysis the chaotic characteristic parameters of PD time series of the oil-paper insulation at different voltage, it has been found that the largest Lyapunov exponent of attractors decrease, correlation dimension and Kolmogorov entropy increase with the increase of applied voltage.⑤The degradation of oil-paper insulation under PD condition is divided into five stages, and BP artificial neural network of which the chaotic characteristics parameters of PD is used as the input is introduced to assess the developmental degree of oil-paper insulation defect. Results show that BPANN could be used to recognize the partial degradation stage of oil-paper insulation in a given error range and L-M algorithm is a much more reasonable BP network algorithm to diagnose the degradation of the oil-paper insulation under PD condition. The lowest recognition rate of five degradation stage is 70 percent, and most is 80 percent or more.