Pressure vessel state investigation based upon the least squares support vector machine

In view of the remarkable time–frequency property obtained from wavelet packets and the excellent generalization ability derived from the least squares support vector machine (LS SVM), a novel approach is proposed, which focuses on the research on state detection for pressure vessels. The minimum entropy criterion is adopted to realize the optimal wavelet packet decomposition, the feature vectors being established according to the percentage of single-band signal energy in the total energy. In addition, the LS SVM is introduced to accomplish classification, for judging the states of pressure vessels. The test results show that high classification accuracy is achieved compared with the cases for the original SVM and BP neural networks under the same conditions. The scheme proposed is proved to be an accurate one for identifying the various states, which can be adapted to wide practical applications.