Principal Component Feature Extraction And Description From Noise Sequence Based On Wavelet Packet Analysis

Application of wavelet packet transform in feature extraction of time series has obtainedbetter effect, and been paid wide attention. Based on the wavelet transform, this paperintroduced a half soft threshold and improved it for signal de-noising. Then, the characteristicof logarithmic energy entropy got from wavelet packet coefficients are calculated. A scorevector could be obtained from the characteristics vector which is analyzed by principalcomponent analysis (PCA). It can be the foundation of later analysis, modeling andforecasting of time series. The main content includes:1) Noise reduction pretreatment of time series. In the process of de-noising, a new halfsoft threshold method based on accelerated chaotic search is proposed. Select a half-softthreshold function containing adjustable parameters. Then the cubic mapping chaotic searchmethod, which has advantages as good rapidity, ergodicity and so on, is introduced to globallysearch the adjustable parameters of the selected semi-soft threshold function within valuerange. Finally obtain the best estimate wavelet coefficients between the hard threshold andsoft threshold. Using root mean square error (RMSE) and signal to noise ratio (SNR) toevaluate the effect of de-noising, simulation results show the effectiveness of the method inthis paper.2) Feature extraction of time sequence. Logarithmic energy entropy is a kind ofparameter which can describe confusion degree of the system. Based on the wavelet packettransform, the wavelet packet coefficient of multiple scales is calculated. Then, the featureextraction of logarithmic energy entropy can be calculated. According to the nature of thelogarithmic energy entropy, the state of the system can be judged.3) Health detection of the system state based on the characteristic parameters. As the toolsignal for example, the feature of the different period sampling time sequences is extracted.Principal component scoring map is gotten by the application of PCA. According to thecontrol limit in the diagram, it can tell the current health state of system reflected by time sequence.4) Design and development of principal component feature extraction system from noisysequence. Design and development of system using VC6.0platform and Matlab7.1platformincludes three functional modules, which is feature extraction preprocessing, featureextraction and feature parameters modeling. The system interface can fully demonstrate thewhole process of the feature extraction of the sequence with noise.