Power transformer vibration signal fault diagnosis based on feature determination coefficient

Abstract

Transformer live fault diagnosis is of great significance to ensure the safe and stable operation of power transformers. In response to the problem of complex working environment and limited fault types characterized by a single parameter, a method based on complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) and feature entropy weights method (EWM) is proposed for fault diagnosis. The correlation coefficient weighted kurtosis (CCWK) principle is used to filter the CEEMDAN components and reconstruct the signal to achieve an improved characterisation of transformer vibration signal features while eliminating redundant components. The EWM is used to construct feature determination coefficients (FDC) to achieve a single data diagnosis of transformer fault types. The principal component analysis (PCA) is used to reduce the scale of mixed domain features and the chicken swarm optimization (CSO) algorithm is used to optimize support vector machine (SVM) model for fault diagnosis. The analysis is performed on a 110 kV three-phase oil-immersed transformer in a certain substation, and the results show that compared with other transformer fault diagnosis methods such as probabilistic neural network (PNN) and SVM, the proposed method not only provides early qualitative fault type identification but also improves the accuracy and efficiency of transformer fault diagnosis.

Keywords

• fault diagnosis
• transformer vibration signal
• complete ensemble empirical mode decomposition with adaptive noise (ceemdan)
• signal-to-noise ratio
• entropy weight method(ewm)
• support vector machine (svm)
• chicken swarm optimization algorithm