V.I. Marchuk, N.V. Gapon, V.A. Franc

The main task of measurement results processing is to increase the reliability of the data that needs to be addressed such issues as reducing the variance of the random component. It take in account fetching of useful component in the non-parametric a priori uncertainty and limited results. The main difficulty in the separation of the initial implementation of the measurement results of a utility and a random component is the lack of a priori information about the statistics of these processes and the requirement of full automation of its procedures for allocating the components without the participation of the experimenter. In this paper we present a theoretical basis for error estimates of the selection signal with limited a priori information about the function of the useful component and the statistical characteristics of the additive noise component. The expressions obtained allow to speak about the asymptotic optimality of estimates of breeding method, which is confirmed by experimental studies.