Bayesian Unmasking Method To Detection Of Gross Errors

In the field of surveying data processing，in order to ensure the quality of surveyingdata,data should often be dealt with and gross errors should be deleted. However, in gross errorsdetection, both independent observations and correlated observations are facing the problem thatgross errors could not be detected accurately, namely, masking and swamping are happening. Soin this paper, modern bayesian statistic theories will be used to research the problem of maskingand swamping in gross errors detection.With the method detection of gross errors based on posterior probability of observationerror as base, combined with the influence analysis, Bayesian unmasking method to delection ofgross errors together with influence analysis is put forward which is seasoned with independentobservations; Considering the failure of Gibbs sampler and eliminating leverage points withoutliers from the conditional subset, multiple outlier detection in observations including leveragepoints with outliers by Bayesian method is put forward which is also seasoned with independentobservations; With the model change as jumping-off place, combined the method detection ofgross errors based on posterior probability of observation error with the theoretics of reliability,Bayesian unmasking methods for gross errors detection of correlated observations is put forwardwhich is seasoned with correlated observations.Major research achievement and innovation as follows:1. Bayesian unmasking method to delection of gross errors together with influence analysis.By analysis the reason of masking and swamping and combined the method detection ofgross errors based on posterior probability of observation error with the influence analysis based onthe divergence of Kullback-Leiber, a Bayesian method for gross errors detection which is namedBayesian unmasking method to delection of gross errors together with influence analysis isproposed based on idea of one-by-one searching. This method is good at unmasking andunswamping for gross errors detection in independent observations.2. Multiple outlier detection in observations including leverage points with outliers byBayesian method.By considering the failure of Gibbs sampler and eliminating leverage points with outliersfrom the conditional subset, a Bayesian unmasking method for multiple outlier detection whichis simple and easy deal is given. The experiment result shows that this method has the functionof unmasking and unswamping in independent observations.3. Bayesian unmasking method to delection of gross errors in correlated observations. With the model change as jumping-off place, though detailed theoretical explaination and inthe principle of simple, a Bayesian method for gross errors detection which is named Bayesianunmasking method to delection of gross errors in correlated observations is proposed forunmasking and unswamping in correlated observations. For proving the validity of this method,the method is applied to gross error detection in GPS network. It is shown that procedure iseffective and applicable.