The research on the method of condition estimate of power transformer base on support vector machine

The transformer condition estimate model is constructed based on SVM and the parameter for SVM-based classifier is determined by adopting cross validation method. Considering the compactness characteristics of DGA data and combining the own characteristics of SVM, a method of fuzzy C-means clustering to pre-select representative samples is presented. Simulation results show that, the pre-selection of representative samples could solve the problem of time consuming on parameter determination and enable to detect transformer faults with higher diagnosis accuracy. The combination of fuzzy clustering method with SVM is also helpful to other pattern recognition problems.