Multi-kernel support vector classifier for fault diagnosis of transformers

Dissolved gas analysis (DGA) has proved to be one of the most useful techniques to detect the incipient faults of power transformers. This paper presents a novel method named multi-kernel support vector classifier (MKSVC), to analyze the DGA for fault diagnosis of transformers. Different from the conventional support vector machine (SVM), MKSVC uses a combined kernel formed through a linear combination of several basis kernels. In MKSVC, each basis kernel extracts a specific type of information from the training data, providing a partial description of the data. Given many partial descriptions of the data, a convex optimization is obtained by a linear combination. Thus, the learning problem can be solved by iteratively computing this optimization problem. The MKSVC method is evaluated using 318 fault data in comparison with several commonly used methods. The diagnostic results show that the diagnostic accuracy of MKSVC prevail those of the commonly used methods.