Application of DDAGSVM in Fault Diagnosis for Electric Transformer

Electric transformers play an important role in the electrical power system, and there is a strong demand on their reliable and safe operation. Support vector machine (SVM) based classification gives a promising approach for fault diagnostics of electric transformers. But the standard method for N-class SVMs (there are many types of electrical transformer fault) doesn't present an easy solution. The constructing N-class SVMs are still an unsolved research problem. In this paper, we apply a new learning architecture (the Decision Directed Acyclic Graph Support Vector Machine, DDAGSVM) in electric transformer fault diagnosis and try to obtain good results. The DDAGSVM operates in a kernel-induced feature space and uses two-class maximal margin hyperplanes at each decision node of the DDAG. At the last example shows that DDAGSVM is substantially faster to train and evaluate than either the standard algorithm (1-v-r) or Max Wins in electric transformer fault diagnosis, while maintaining comparable accuracy to both of these algorithms.