Abstract: In order to handle the problem that the feature extraction of dissolved gas analysis (DGA) data by principal component analysis (PCA) is not distinct, a new transformer fault diagnosis method based on relative transformation (RT) PCA is proposed. The original data space is converted to the relative data space by relative transformation which makes the transformed data more distinguishable. PCA is employed to reduce the dimension of relative space to make the features more representative in the relative space. Diagnosis model based on least squares support vector machine (LSSVM) is set up according to the fault characteristic of transformer. The chaos particle swarm optimization (CPSO) algorithm is adopted to optimize the kernel parameters of LSSVM. Compared with PCA-LSSVM, RT-LSSVM and grey relation entropy, experimental results show that RTPCA increases the separability of data set and verify the fault diagnosis ability of the proposed method.

Key words: transformer, fault diagnosis, relative transformation, principal component analysis, least squares support vector machine, particle swarm optimization algorithm