Transformer Fault Diagnosis Based On Support Vector Machine

In the wake of social progress and continual growth of national economy, the electric power industry in China reaches fast development with increasing capability in generation, transmission and distribution of electric power. The power transformer, which is responsible to the voltage transformation, the distribution and transmission of electric power and providing customer services, plays a key role in the power system. Whenever malfunction of the transformer occurs, influence will reach a large scale till it is fixed after a long period. Thus, the reliability of the transformer is an essential factor of that of the whole power system, any improvement of which will bring in great benefits of safety, reliability and economic operation to the power system and also to the society.At present, oil-immersed transformer is mainly used all over the world, which is thus focused on in this study. Transformer faults mainly occur in the internal insulation with the characters of badly degradation. The experiences of transformer fault analysis have been accumulated from the growth of power detection technique. Electronic, physic and chemical methods are used to detect and diagnose transformer condition, which, in detail, include dissolved gas analysis (DGA), online detection of pressure, infrared temperature measurement, partial discharge experiment, ultrasound localization, immersed prediction, immerse oil aging experiment, transformer pressurization detection, winding deformation detection, flow electrification in oil detection and some other common detection.The limitation of former relative studies was analyzed and figured out as deficiency of distinguishing only two types of power transformer condition. In this study, the diagnostic model of transformer condition was established on C-support vector machine (SVM) combined with the oil chromatography analysis, which was able to identify multiple conditions of the transformer. The C-support vector classification was constructed to implement identifying different fault or normal conditions by applying the learning savoir-faire of SVM which comes from data-based statistical learning theory and by employing the method of cross-validation to adjust and optimize the major parameters.The model, which possessed high accuracy of classification prediction, can distinguish and diagnose the oil chromatography data sets with high-performance. It is simple and functional with the potentiality of popularization and application, providing a reliable approach for transformer fault diagnosis in practice.