Fault diagnosis for power circuits based on SVM within the Bayesian framework

Based on least squares wavelet support vector machines (LS-WSVM) within the Bayesian evidence framework, a systematic method for fault diagnosis of power circuits is presented. In this paper, the Bayesian evidence framework is applied to select the optimal values of the regularization and kernel parameters of multi-class LS-WSVM classifiers. Also wavelet coefficients of output voltage signals of power circuits under faulty conditions are obtained with wavelet lifting decomposition, and then faulty feature vectors are extracted from the disposed wavelet coefficients. The faulty feature vectors are used to train the multi-class LS-WSVM classifiers, so the model of the power circuits fault diagnosis system is built. In push-pull circuits, this method is applied to diagnose the faults of the circuits with simulation; the results show that the fault diagnosis method of the power circuits with LS-WSVM within the Bayesian evidence framework is effective.