Font Size: a A A

Research On Deformation Analysis Model Based On Wavelet Transform Theory

Posted on:2005-06-30Degree:DoctorType:Dissertation
Country:ChinaCandidate:H Y WenFull Text:PDF
GTID:1100360182467700Subject:Geodesy and Survey Engineering
Abstract/Summary:PDF Full Text Request
Deformation monitoring is a fringe topic in the field of prevention and reduction of disaster in geodesy both at home and abroad. Because of the continual applications of Global Position System (GPS), survey robots and highly advanced technology such as sensor integrated system, deformation monitoring accumulates a great deal of information and datum, it will possess theory meaning and practical value that how to proceed excavating efficiently from abundant information of deformation monitoring data in time, and how to analyse, to interpret and to proceed studying of short-term or medium term or long term forecast in deformation. Deformation analysis generally includes deformation data preprocessing and deformation forecast. This dissertation principally studies the analysis models based on wavelet transform theory. The main contents include:(1) Study data denoising method by wavelet filter of deformation data preprocessing.(2) Study multiresolution (multiscale) deformation analysis models based on wavelet transform.(3) Study wavelet neural networks (WNN) deformation forecast method.As for noises different distribution of deformation analysis data preprocessing, this dissertation analyses the characteristics of wavelet transform, puts forward improved extending algorithm principle of denoising by wavelet decomposition and reconstruction and new algorithm of threshold value method by nonlinear wavelet transform, provides estimation method of root mean square error (RMSE) and adaptive algorithm of threshold value, proceeds analyzing theoretically such characteristics as vanishing moments, regularity, compact support and symmetry of wavelet function, makes experiments in testing practical application effectiveness while a deformation observation time series containing Gaussianing noises, variety of system interfering signals and singularity signals with several wavelet function, studies how to select optimal wavelet basis in deformation observation data preprocessing. The dissertation establishes multiscale wavelet autoregressive deformation analysis model, wavelet multiscale Kalman filter model and its algorithm, wavelet neural networks model and its algorithm. The wavelet multiresolution (multiscale) is integrated organically with Kalman filter and the strongly approaching ability of artificial neural network, which exerts their respective predominance, forecasts the nonlinear deformation, thus provides a new means and method for deformation analysis and deformation forecast. The dissertation includes six chapters.The first chapter is an introduction. It summarizes the actuality and development related with deformation analysis and forecast models. It concludes that different mathematical models have their application characteristics respectively due to different research angles, starting point of model building, adoptive data form, sub sample size as well as applied condition. It makes comment on their limitation, overviews the actuality and the future research & development trend of deformation analysis and forecast, discusses the progress of wavelet analysis in theoretical research and application and its studying actuality in surveying and mapping field, points out subsistent questions, and puts forward the research content of this dissertation.The second chapter is the research of wavelet analysis and its application. This chapter expounds the development from Fourier transform to wavelet analysis including orthogonal wavelet and wavelet time series. The multiresolution analysis concept and Mallat algorithm of discrete wavelet decomposition and reconstruction are introduced. Several kinds of typical wavelet function in common use are introduced. The chapter expounds the evolution of wavelet analysis, combining this text with the research content of the dissertation, discusses the research practicability by usingwavelet to deformation data preprocessing, to make deformation analysis models and proceed deformation forecastThe third chapter is the study methods of wavelet filter and denoising. The research of wavelet decomposition and reconstruction principally focuses on the research maximum scale determining, edge treatment and extending algorithm in denoising. In denoising by wavelet threshold value methodf the determination of estimation method of root mean square error(RMSE) and improved adaptive algorithm of nonlinear wavelet threshold value have been studied and a kind of threshold improvement smoothing algorithm of denoising by nonlinear wavelet transform is put "forward. Experiments are made on optimal wavelet function selection in wavelet filter in application effect of deformation data denoising. The different wavelet functions are compared and analyzed while a deformation observation time series contains Gaussianing noises, variety of system interfering signals or singularity signals. The wavelet function characteristics of denoising in the above condition are revealed and reference to how to select optimal wavelet function in deformation data denoising are providedThe fourth chapter is the study of wavelet multiscale model in deformation analysis. Wavelet multiscale characteristic and correlated characteristic of observation time series after wavelet multiple scale transform are discussed and observation time series covariance after wavelet multiple scale transform are derived. Put forward wavelet multiscale time-frequency analysis method, separate trend component with periodicity component from deformation observation time series, and establish fitting model respectively: A good result is obtained in deformation forecast by this method. Based on the building principle of wavelet multiscale autoregressive model, the wavelet multiscale autoregressive frame and wavelet multiscale autoregressive model are discussed, Kalman filter model of autoregressive forecast and its algorithm are studied and discrete wavelet multiscale wavelet Kalman filter model is put forward. Comparing the practical examples of deformation observation data processing, the results of wavelet multiscale Kalman filter model are better than those of single method of wavelet denoising or Kalman filter in enhancing and improving data precision of real time and dynamic deformation observation. The effect of precision- improving is notable. Theory formulae of wavelet multiscale Kalman filter errors are derived in detail. The researched result can be used to direct selection suitable model of noise variance matrix, which can bring designed wavelet multiscale Kalman filter model proceeding computation in stated range of error.The fiftti chapter is the application of wavelet neural networks in deformation forecast. It summarizes the research actuality and development of artificial neural networks and wavelet neural networks models, reviews the nonlinear time series model and wavelet network structure model combining affine transform with rotation transform of wavelet neural networks, sums up the computation procedure of wavelet neural networks model and take Morlet wavelet function as an example, derives gradient algorithm of wavelet neural networks model, then puts forward a kind of wavelet neural networks deformation analysis forecast model with a case in its application. The sixth chapter is the conclusion and outlook in which the major work and innovative points of the dissertation are summarized, including some suggestions and prospect for the research in the future.
Keywords/Search Tags:wavelet analysis, multiscale, Kalman filter, wavelet neural networks, deformation forecast
PDF Full Text Request
Related items