МЕТОД СОПРЯЖЕННЫХ ГРАДИЕНТОВ В СИСТЕМЕ АВТОМАТИЧЕСКОГО ПОСТРОЕНИЯ ПРОГНОЗИРУЮЩИХ МОДЕЛЕЙ

Abstract

PURPOSE. The paper deals with the increase in reproduction accuracy of time and frequency units through the use of predictive autoregressive-moving average models (ARMA)). The methodology of ARMA model creation existing now is based on interactive procedures. This requires highly qualified specialists and prevents the algorithms of optimum filtration from the introduction in the practical activities of time services. METHODS. The study employs the methods of time series analysis, ARMA model construction and the conjugate gradients method. RESULTS. An approach allowing a complete formalization of the procedure of model construction is introduced. A program module implementing the automatic construction of the predicting models that describe the processes of hydrogen standards frequency variation is developed. It is tested experimentally and the results of the work confirming the adequacy of the obtained models are presented. CONCLUSIONS. The formalized methodology of ARMA models construction proposed by the authors will allow to solve the problem of full automation of time series model construction by empirical data and to lower the reproduction error of time and frequency units by group standards up to 30%.The conducted study provides all the reasons to suppose that the developed software module can serve as a basis for creating a standard software for the subsystem of internal comparisons of time and frequency standards in order to be introduced into the experimental operation mode.