Structural-Parametric Adaptation of Multilayer Information Processing Systems Using Local Quality Functionals
Keywords:
Adaptation, System of Measurement Information Collection/Processing, Multiparameter Optimization, Neural Nets, Multilayered MLP-QualifierAbstract
Purpose: A wide class of application-oriented tasks deal with reallocation of restricted resources in hierarchical systems to
provide extremes for the output quality coefficients, when the functionals of the internal level elements quality are restricted.
If these tasks have big dimensionality, there are methodological problems of developing optimal models of distributed informationprocessing systems which function under the conditions of nonstationary changes of the local quality functionals. The aim of this work is developing a model and a method of structurally-parametrical adaptation of hierarchical big-dimensionality systems, localizing the changes of structural elements quality functionals. Results: Application of advanced mathematical apparatus of optimization methods combined with the error backpropagation algorithm in multilayer MLP-networks gave rise to the model in which the quality parameters of the measurement information collection/processing system are adapted to the structural and/or parametric changes of the information. The problem was formulated of synthesizing an input/ouput multidimentional adaptation model for a distributed system of measurement information collection/processing. Taking into account the similarity of hierarchies of the modelled processes and the architectures of the homogeneous computing environments, the modelling used the multilayer MLP-qualifier architecture. The method of structural-parametric adaptation of hierarchical systems was developed, with the usage of structural elements quality functionals. Practical relevance: The model is useful in defining the adjustments of functional elements at various hierarchical levels, proceeding from the given quality functional for the whole system at the top level of its hierarchy. The dependences were obtained which help to select an algorithm for the parametric synthesis of a hierarchical system information structure model, depending on the state of a set of its elements.