Interpretation of DGA for transformer fault diagnosis with complementary SaE-ELM and arctangent transform

This paper presents a novel approach for power transformer incipient fault diagnosis through the analysis of dissolved gas in oil. The proposed approach is implemented for improving the diagnosis accuracy by dissolved gas analysis (DGA) of power transformer based on the combined use of a multi-classification algorithm self-adaptive evolutionary extreme learning machine (SaE-ELM) and a simple arctangent transform (AT). On the one hand, the SaE-ELM algorithm has the ability to approximate any nonlinear functions with its structure parameters, i.e. hidden node biases and output weights, optimized self-sufficiently. On the other hand, the AT can alter the data structure of the experiment data, which will enhance the generalization capability for SaE-ELM as well as other machine learning algorithms. Thus, the combination of SaEELM and AT can complement each other and improve the diagnosis accuracy from the aspect of both algorithm and data structure. The performances of the proposed approach are compared with that derived from ANN, SVM, and ELM methods, respectively. Experimental results with both published and power utility provided data indicate that the developed approach can significantly improve the accuracies for power transformer fault diagnosis.