System Structure Identification by Sets of Observation Data on the Base of Unbiasness Principle

Abstract

Phisical model for identification of the complex system and nonphisical models for prediction of processes can be obtained by sifting of many models-candidates by external criteria. Having the small input data sample available the difference type criteria of non-contradiction and balance of variables are the best for sifting (or selforganization) of models. The special procedures for criterion regularization and for unimodality of criterion characteristic are considered. These procedures are provided by the algorithm of Objective System Analysis (OSA) where the systems of equations are sifted.Then iterative Convergent Multilayered Linear (CML) algorithm provides the optical models in the form of system of equations systems obtaining. The computation time is evaluated. The unimodality of characteristic is provided also by algorithm of Objective Computer Clasterization (OCC) which have the similar procedure of clasterizations sifting (Ivakhnenko A.G., L.P. Semina, and G.A. Ivakhnenko, 1987).