Application of GMDH type neural network for predicting UTC(k) timescales realized on the basis of hydrogen masers

The article presents research results on predicting the deviations for UTC(AOS) timescale realized on the basis of hydrogen maser by means of GMDH type neural network. Input data are prepared in the form of two time series (TS1 and TS2). Time series TS1 is built on the basis of values of comparison between the UTC(AOS) timescale with a clock realizing this scale, and values of deviations determined according to UTC and UTC Rapid scales, published by the BIPM. Time series TS2 is built only on the basis of values of deviations determined according to UTC and UTC Rapid scales. Better quality of predicting the UTC(AOS) timescale is obtained for data prepared in the form of time series TS2. Obtained values of predictions differ from the deviations published by the BIPM at the same day of prediction by ±10 ns for time series TS1 and ±5 ns for time series TS2.