A transformer fault diagnosis method based on sub-clustering reduction and multiclass multi-kernel support vector machine

The application of big data technology to the equipment asset management helps to enhance the operational reliability of the energy system. This paper is focused on transformer fault diagnosis using support vector machine (SVM), one of the most commonly used data-mining methods. The direct multiclass SVM model is capable of obtaining a single classification function and avoids the difficulties of designing multiple groups of model parameters. However, when dealing with mass samples, the direct model will lead to low training efficiency and curse of dimensionality. To make the direct multiclass SVM-based transformer fault diagnosis not subject to the sample size limit, the sample reduction algorithm based on K-medoids clustering is introduced. Before multiclass SVM training, potential support vectors can be extracted from the large-scale training set via sub-clustering method, which improves the calculation efficiency. Multi-kernel learning is also applied to the SVM model to further boost the classification performance. The effectiveness and application prospect of the proposed method are verified in case study.