Fault classification method for inverter based on hybrid support vector machines and wavelet analysis

A new classification method for fault waveform is proposed based on discrete orthogonal wavelet transform (DOWT) and hybrid support vector machine (hybrid SVM) for fault type of a three-phase voltage inverter. The waveforms of output voltage obtained from the faulty inverter are decomposed by DOWT into wavelet coefficient matrices, through which we can obtain singular value vectors acted as features of time-series periodic waveforms. And then a multi-classes classification method based on a new Huffman Tree structure is presented to realize 1-v-r SVM strategy. The extracted features are applied to hybrid SVM for determining fault type. Compared to employing the structure based on ordinary binary tree, the superiority of the proposed SVM method is shown in the success of fault diagnosis because the average Loo-correctness of the SVM based on Huffman tree structure exceed the general SVM 3.65%, and the correctness reaches 99.6%.