Research On Transformer Fault Diagnosis Based On Gas Dissolved In Oil And Ultra High Frequency Partial Discharge Signal

Transformer fault diagnosis problem has a long history, as an important transformer equipment, its stable operation is related to the stability of the power supply. How to reflect the fault characteristics as the fault itself is the key, the two are non-linear relationship. Their predecessors have fault characteristic of oil and gas combined with artificial intelligence algorithm on this issue has made certain achievements, formed a set of mature based on fault diagnosis theory of oil and gas. With the continuous development of transformer diagnosis technology, audio signal, image information, electrical signal, oil leakage and temperature have become the characteristic information of fault diagnosis, a new fault diagnosis method is formed.In the actual conditions, because the fault is complex, the single feature information can not be fully transformer fault diagnosis, some of the characteristics of the information can be qualitative fault, and some are suitable for fault location, therefore, the traditional single source information diagnosis technology can not meet the needs of transformer fault diagnosis.Transformer fault diagnosis based on oil volume is an important research content in this paper. Transformer fault or abnormal operation, will produce the various features information of gas dissolved in the oil. This diagnosis method for transformer fault detection and fault qualitative has a very good effect, In this paper, we use the BP neural network theory which has the nonlinear problem processing ability, and the adaptive mutation particle swarm optimization is used to optimize the BP network, and the simulation results are implemented in MATLAB.The partial discharge signals in transformer diagnosis, this method can more accurate fault location using ultra high frequency signal based on, according to the measured waveform data can be roughly determined discharge type. From this kind of feature information mechanism, and the specific positioning principle do related elaboration, for several kinds of typical partial discharge were modeling and simulation, In this method, wavelet packet decomposition and kernel principal component analysis are used to study the feature extraction and optimization.Finally the concept of information fusion was related in this paper, introduced some fusion model and algorithm. In order to improve the diagnostic efficiency, the two methods are combined. the decision level fusion, it has high flexibility, fault tolerance, strong anti-interference ability, and the dependence of the single sensor is small.