The nature of statistical learning theory
The nature of statistical learning theory
A comparison of methods for multiclass support vector machines
IEEE Transactions on Neural Networks
Hi-index | 0.01 |
Aiming at to the characteristics of fault diagnosis of analog circuit with tolerances, noise, circuit nonlinearities and small sample set, this paper presents a novel algorithm for the existent problems based on incomplete wavelet packet transform for feature extraction and improved balanced binary-tree support vector machines (BBSVMs) for multi-classification. Firstly, analyzing characters of the optimal wavelet packet transform and the incomplete wavelet packet transform, the conclusion that the latter is perfect for fault diagnosis of analog circuit is obtained. Secondly, in order to perform multi-class classification, binary-tree-based SVMs has been studied. By introducing the separability measure that based on the space distribution of pattern classes, an improved balanced binary-tree structure is constructed. The proposed algorithm for analog circuits with tolerances and noise is implemented, and simulation results show us that compared with several existent fault diagnosis methods, the current algorithm has highest classification speed and higher classification accuracy, which make fault diagnosis of analogy circuit on-line promising.