Research On Power Transformer Fault Diagnosis Method Based On Improved ISOMAP And WkNN

As a very important power equipment in the power system,power transformer undertakes the core work of power transmission,conversion and distribution.Once it breaks down,it often brings huge economic losses,even casualties and power system paralysis.In order to improve the power supply reliability of power system,it is necessary to ensure the safe operation of power transformer.Although the traditional diagnosis methods can diagnose the power transformer fault preliminarily,they still have many defects and the accuracy of diagnosis is limited Therefore,in order to diagnose the potential faults of power transformer timely and effectively,the research of intelligent fault diagnosis methods has important practical significance.In this paper,power transformer fault is studied from three aspects: data preprocessing,feature extraction and fault diagnosis:(1)The preprocessing methods of power transformer fault data are studied.Aiming at the problem that the distribution of fault data of power transformer is quite different in order of magnitude,this paper makes a comparative analysis of various data preprocessing methods,and finally selects LOG method as the preprocessing method.This method is a non-linear data preprocessing method,which can make the transformer fault data more evenly distributed in the space,so as to reduce the adverse effect of the bad data distribution on the algorithm.(2)A supervised Isometric Feature Map(ISOMAP)algorithm based on sample evaluation is proposed.In order to extract the important information contained in transformer fault data effectively,the algorithm of ISOMAP is improved,and a supervised algorithm of ISOMAP method based on sample evaluation is proposed.Firstly,according to the membership degree of each sample point to each fault category,the reliability of each sample point is evaluated,and the sample label information is used to reconstruct the geodesic distance matrix,and through multidimensional scaling,the low-dimensional embedding vectors are obtained.Then,RBF neural network is used to construct the mapping relationship between the original space and the low-dimensional space to reduce the dimension of testing datas,so as to obtain the extracted fault features.Finally,the fault diagnosis example shows that the method can effectively improve the separability of fault categories.(3)An adaptive k-value Weighted k-Nearest Neighbor(WkNN)algorithm is proposed and applied to power transformer fault diagnosis.In order to improve the defect of fixed k-value in traditional kNN algorithm,an adaptive k-value WkNN algorithm is proposed.Firstly,the k-value is adaptively selected according to the local density of the sample datas,and the Euclidean distance is combined with the distribution similarity of the sample datas to determine the weights of the neighbor points,so that the fault categories of the testing datas can be classified more effectively.Then,the power transformer fault diagnosis is carried out by combining the improved ISOMAP algorithm and the adaptive k-value WkNN algorithm.Finally,the fault diagnosis example shows that the proposed method can further improve the accuracy of transformer fault diagnosis.