Research On Detecting Methods For Outliers In GNSS Time Series And Its Applications

Effective and reliable GNSS data is the premise and foundation for precise positioning,navigation and timing.The outlier detection in GNSS time series is an important part of improving data reliability.This article is mainly devoted to the research of GNSS time series outlier detection methods and it's applications.Based on the ARIMA model,the Bayes method,likelihood ratio method,EM algorithm and model selection method of GNSS time series outlier detection are proposed,and they are applied to satellite clock offset data processing and the detection and repair of BDS satellite tri-frequency cycle slip.The main work and contributions of this article are as follows:1.A Bayes method of GNSS time series outlier detection is proposed.Using the theory and method of Bayes statistics,from the perspective of Bayes hypothesis testing,the detection model and discrimination rules of GNSS time series outliers are proposed based on the posterior probability of identification variables;How to select the property prior distribution of ARIMA model parameters and identification variables are studied respectively from different points;the posterior probabilities of identification variables are calculated by Gibbs sampling algorithm.A new satellite clock offset prediction model and outlier detection model are constructed through combining the quadratic polynomial model and the ARIMA model;four different types of GPS satellite clocks are selected randomly,and their clock offset data are used to investigate the effectiveness of the new algorithm.And the detection results are compared with the detection results of MAD method.2.Two new likelihood ratio methods for detecting the outliers in GNSS time series are proposed.Based on the idea of variance expansion model,the abnormal disturbances in data are classified into the random model.Using the likelihood ratio principle and method,an outlier detection model and test rules are constructed,and the outlier detection problem is transformed into a hypothesis test problem;a likelihood ratio method of GNSS time series outlier detection is proposed and the Score test statistics of likelihood ratio test is derived simultaneously.In view of the problem that masking and swamping are prone to occur during the detection of additive outlier patches,so the mechanism how the outlier patches disturb outlier detection is analyzed,and the mechanism how the difference and inverse difference impact outlier detection is analyzed simultaneously,then a new algorithm for detecting additive outlier patches is proposed and a method for estimating abnormal disturbances are proposed.And the new method is applied to process the BDS satellite clock offset data.3.An EM algorithm and two improved EM algorithms of GNSS time series outlier detection are proposed.The identification variable is introduced to establish an outlier detection model based on the ARIMA model,and the identification variable is regarded as a hidden variable.The EM algorithm is used to determine the locations of the outliers and calculate the magnitude of abnormal disturbances.To solve the problem that the coefficient matrix of EM algorithm is prone to ill-conditioned in the GNSS outlier detection process,the EM algorithm of outlier detection is improved by using biased estimation theory and regularization methods,respectively.The corresponding determination schemes of biased parameter and regularization parameter are given.And those methods are applied to process the data of GPS and BDS satellite clock offset.4.A model selection method of GNSS time series outlier detection and the two-stage method of outlier patches detection are proposed.From the perspective of model selection,the GNSS time series outlier detection model is established,and the detection problem of outliers is transformed into a model selection problem.The MDO measurement standard of GNSS time series outlier detection is proposed to solve the problem how to determine the location and the magnitude of outlier;the two-stage method of outlier detection and the criteria for outlier determination are proposed;the new algorithm is applied to process the data of GNSS satellite clock offset.And the results are compared with the results of common outlier detection methods under the indicators of RMSEP,Mean and MAB.5.Those new outlier detection methods are applied to the detection and repair of BDS satellite triple-frequency cycle slip.To solve the problem that the triple-frequency geometry-free phase combination is insensitive and the data correlation is enhanced by multiple combinations,the GNSS time series outlier detection methods proposed in this paper are applied to the detection and repair of the BDS triple-frequency combination cycle slip.Through the detection and repair experiments of isolated cycle slip,continuous cycle slip,random cycle slip of different satellites and multi-satellite multi-site combined cycle slip,the effectiveness and reliability of the four new methods for cycle slip detection are verified.