Study On State Monitoring And Fault Diagnosis Of Transformer Winding And Core Based On Vibration Analysis

With the rapid growth of power capacity, the requirements for the safe operation and the reliability of power system becomes more and more urgent. Transformer's safe operation is of great significance to the entire social because of its value and vital social means, while its winding and core failure was one of the major faults, accounting for a considerable proportion in the overall failure. Implement of state monitoring and fault diagnosis of transformer winding and core is the precondition of predicting maintenance, is the key element of reliable run, and is the important supplement and updated development to the traditional preventive maintenance.This thesis investigates methodology of the state monitoring and fault diagnosis of transformer winding and core. First of all, the Hilbert-Huang transform, BP neural network and least squares support vector machine are deeply studied. After that, they are applied into fault diagnosis system of transformer winding and core. At the same time, the winding and core vibration is studied to analyze the relationship between the vibration signal and their conditions. To cope with the problems raised above, the research is conducted. The main work and conclusions are correspondingly listed below:The main reasons and the pipelines of the transformer vibration is analyzed. And the winding is predigested as a mathematical model, theoretically analyzing the changes of the pre-compressions on windings. On the other hand, the principle of core magnetostriction and its influencing factors are presented. Once the core's compression decreases, temperature varies or the insulating layer gets scathing, the core vibration signal will change. Therefore, from the core vibration signal the core fault can be monitored.According to the characteristics of transformer vibration signals, this thesis designed a set of on-line monitoring system, and the Hilbert-Huang Transform is applied into the vibration signal analysis, and gives the specific process of the algorithm. For the noise interference problem, a new method based on IMF adaptive thresholding processing is investigated. Compared with the conventional wavelet-based denoising algorithms, the new algorithm is simpler, more flexible, and not limited by the selection of wavelet function and optimal decomposition level of wavelet, meanwhile, automatic selection of threshold value and thresholding layers of intrinsic mode functions are realized.It is difficult to distill signal characters for the fault diagnosis. With the character of vibration signal, the thesis gives one algorithm to distill signal characters. The algorithm is based on wavelet packet transform and information entropy theory, and it can transform signal characters into digital characters. The algorithm is the basis of the fault diagnosis.At the end of this thesis, BP neural network and least squares support vector machine(LSSVM) which optimized by improved particle swarm optimization (IPSO) are used to forecast the fault diagnosis of transformer winding and core. Case study show that the IPSO-BP and IPSO-LSSVM algorithms can obtain satisfied classification result, and diagnosis speed and accuracy by IPSO-LSSVM algorithm is better than IPSO-BP algorithm. Consequently, the IPSO-LSSVM algorithm is a proper alternative for fault diagnosis of transformer winding and core.