Chaotic Characteristics Extraction And Pattern Recognition Method For Partial Discharge

Electrical insulation plays a very important role in the operation of the high-voltage electrical apparatus,insulation condition is directly related to the reliability of electrical apparatus.Effect of insulation deterioration is mainly due to dielectric internal partial discharge,and therefore the characteristics of partial discharge signal analysis and pattern recognition is significant to the on-line monitoring and fault diagnosis of partial discharge development and insulation level of internal defects.The main reason of partial discharge is the nonuniform distribution of electric field in dielectric.In this dissertation,five partial discharge models based on the degree of non-uniform electric field distribution are built,using the pulse current method to obtain the maximum partial discharge capacity of different models with time diagram and the three-dimensional spectrum.Then the phase space reconstruction and chaotic attractor and chaotic characteristic parameter extraction of time series of the maximum partial discharge are performed by using chaos theory,after that,the phase space reconstruction and chaotic characteristic parameters are considered as a characteristic quantity to characterize the results of chaotic analysis of partial discharge(CAPD).Afterword,the CAPD characteristic quantity is recognized by back propagation(BP)neural network,and recognition results are compared with the results of conventional fingerprint characteristic quantity.Finally,the CAPD characteristic quantity and fingerprint characteristic quantity are combined to be recognized by BP neural network.The results show that the chaotic attractors and the maximum Lyapunov exponents,Kolmogorov entropy and chaos correlation dimension characteristic parameters,which are extracted by CAPD of partial discharge time series indicates that partial discharge of gap is not a completely random process,but rather a chaotic process.Thus the characteristics of partial discharge can be analyzed by the Chaos theory.Besides,the CAPD characteristic quantity and fingerprint characteristic quantity can be recognized better by BP neural network,and recognition results of chaotic characteristics has obvious advantages in terms of needles-needles and ball-electrode plate,but the overall recognition rate close,so the two kinds of characteristic quantity can be complementary.Finally,the recognition rates of comprehensive characteristic quantity are improved for all types of discharge,and the recognition rates of the five patterns can reach up to 92.5%.