Research On Feature Extraction And Prediction Methods For GNSS Deformation Sequences

Global Navigation Satellite System(GNSS)is widely used in many fields such as deformation monitoring because of its simple,efficient,all-weather observation and high precision.For small-scale deformation monitoring,the differential technique can effectively attenuate system errors such as satellite clock error,ionospheric error,tropospheric error and receiver clock error.GNSS monitoring eoordinate sequence is ouly affected by multipath error and random noise,it is necessary to study its error.Features and weakening techniques to improve the accuracy of deformation monitoring,and then scientifically explain it.In this paper,the GNSS deformation sequence is taken as the research object,and the multi-path error extraction and feature analysis and real-time extraction of deformation information are taken as the target.The related research is carried out.The specific research contents and conclusions are as follows:1)Introduce three major GNSS time series denoising methods that are widely used:empirical mode decomposition(EMD),wavelet(WAVELET),blind source separation(BSS),and analysis of three different methods in the GNSS time series denoising process.Advantages and problems.Introducing the latest results of EMD noise reduction method,the improved adaptive noise complete empirical mode decomposition(ICEEMDAN),through simulation and comparison experiments,it shows that the ICEEMDAN method in EMD and its improved algorithm has modal aliasing and endpoint effects in the noise reduction process.Larger improvements;different wavelet bases and wavelet thresholds are used to find the most reasonable wavelet base and wavelet threshold for wavelet threshold denoising,which indicates that wavelet basis haar wavelet and decomposed layers:3 can be used to balance wavelet denoising effect and computational efficiency;The separation method mainly chooses two methods of independent component analysis(ICA)and eigenvalue decomposition(EVD).The ICA noise reduction method has better noise reduction effect and the EVD noise reduction effect is more stable.2)Based on the advantages of three different noise reduction methods,the EMD-WAVELET-BSS coupling model is proposed,which makes full use of the adaptive decomposition ability of EMD method to decompose the time series into a series of intrinsic mode functions(IMF).The high frequency and low frequency parts of the IMF are extracted,and then the wavelet method is used to reduce the noise of the high frequency part and the blind separation and noise reduction of the low frequency part by the blind source separation respectively denoise the high frequency and low frequency parts of the decomposition signal.The noise separated by different methods is used as the second channel to Perform two-channel time series denoising.The simulation experiment and the comparison of measured data show that the ICEMDAN-WAVELET-ICA method is better in GNSS time series noise reduction.3)Analyze the main characteristics of multipath error in GNSS time series?use the diural repeatability of multipath,establish a stellar day filter multipath model in the coordinate domain,and establish a multipath model of Global Positioning System(GPS)system.Research on support vector machine(SVM)in deep learning,establish a predictive network model based on genetic algorithm support vector machine(GASVM)parameter optimization,and predict GNSS multipath error time series.The multi-path prediction model established by the stellar-day filtering multi-path prediction model and the support vector machine shows that the multi-path error prediction model of the GPS single-system support vector machine is basically equivalent to the prediction model of the stellar-day filtering error model;the multi-path of the GNSS multi-system It is difficult to determine the repeating characteristics of the multipath error diural repeatability,but the GASVM model for multipath error can still be established.The analysis of the statistical results shows that the multi-system multi-path model is less accurate than the single-system multipath error modeling,but the prediction accuracy is reached expectations.4)On the basis of the actual measurement,the deformation information is added to the GNSS time series,and the GASVM is used to construct the prediction model of the GNSS time series under the steady state of the network.The tectonic simulation experiment adds the deformation sequence to the stationary time series,and predicts the development trend of the deformed GNSS time series by using the prediction model of the GNSS time series in the stationary state,compares the residual results and the deformation sequence of the GNSS time series,and verifies the prediction deformation.The way of residual prediction of the feasibility of deformation information.Figure[29]Table[9]Reference[89]...