Fault diagnosis of power transformer using kernel-based possibilistic clustering

Dissolved Gas Analysis (DGA) of power transformer oil is an important technique to detect the incipient faults. Recently various artificial intelligence methods have been developed to interpret DGA results such as Artificial Neural Networks (ANNs), Expert System and Clustering Analysis. Against the deficiencies associated with the constrained memberships used in original fuzzy c-means clustering algorithm, the possibilistic c-means clustering algorithm is introduced. Its memberships may be interpreted as degrees of possibility of the samples belonging to the classes. Furthermore, the kernel-based learning method can nonlinear map samples in the original low-dimensional space to a high-dimensional feature space. Then the useful features can be effectively exacted and enlarged for improving the accuracy of clustering. Therefore, a kernel-based possibilistic c-means clustering algorithm is proposed in this paper. The new algorithm is used to analyze DGA data in power transformer. Simulation results are given to illustrate that this algorithm is accurate in clustering and is fast in convergence speed, and it is highly robust in noisy environments.