Study On Analysis Method Of Deformation Time Series Data Based On Wavelet Packet Transform

WPT (Wavelet Packet Transform) is an effective time-frequency analysis method. WPT is used to analyze deformation time-series data in this paper and that supplies an effective analysis method mordern deformation analysis. The contributions are summarized as follows:Wavelet packet threshold de-noising algorithm are used to denoise deformation time-series data. The algorithm with entropy of SURE to choose the best wavelet packet basis, criteria of Birge-massart to estimate threshold and soft-threshold method to quantify coefficients proves to obtains the best denoising effect by simulation experiments. Detect the systemic interference and abrupt changing interference by WPT. Eliminating the systemic interference and using the data around abrupt changing points to interpolate the value of abrupt changing points to inhibit both the impacts of interference. SNR of signal is increased and RMSE is reduced after the pretreatment.The practical wavelet filter is non-ideal so that each sub-band contains frequency components belonging to neighboring sub-band, and down-sampling and up-sampling low-frequency sub-bands with high grequency components and high-frequency sub-bands with 2 causes frequency folding and frequency interleaving, for not meeting the sampling theorem. Measures, reordering the nodes and improved single sub-band reconstruction algorithm, are taken to improve WPT algorithm to weaken the aliasing and effectiveness in eliminating aliasing of the improved algorithm is proved in simulation example.Use FFT to analyze the main frequency components ,and take the improved WPT algorithm to extract frequency bands containing the main frequencies, is brought forward to extract the residual feature of GPS deformation time-series data, and residuals with annual cycle, semi-annual cycle, monthly cycle and semi-monthly cycle are gained with it. Compared to the feature extraction method of applying the classical WPT to detect residuals on each band at each level, computational complexity of the new one is reduced, and residual items acquired by it containing no redundant frequencies, are more believable. That can serve to the evaluation of GPS error correction model and selection of model parameters.