OUTLIER FILTRATION IN PROBLEMS OF STATIC AND DYNAMIC DATA PROCESSING IN TIME AND FREQUENCY STANDARDS

Abstract

PURPOSE. Filtration of anomalous measurements (outliers) of frequency values of hydrogen standards which form a part of the group standards of time and frequency will allow to reduce an algorithmic estimation error of the state vector of group standards. METHODS. The study employs the methods of noise-immunity estimation in the mode of static and dynamic data processing. RESULTS. It is proposed to use α-trimmed estimates with an unknown percentage of outliers to filter the outliers in the static mode. In the case of dynamic data processing, the outliers are rejected if the estimates obtained at this stage go beyond the confidence intervals. The latter are calculated on the basis of the predictive models of frequency variation constructed at the stage of static processing of data. The proposed algorithms are implemented in Mathcad 15.0. The efficiency of the proposed methods has been verified by statistical modeling and confirmed by the processing of real data obtained as a result of VET 1-5 secondary standard operation. CONCLUSIONS. The received results form an important development stage of a formalized estimation procedure of the state vector of the time and frequency group standard in static and dynamic modes based on the use of predictive models.