The article describes the practical application of information-statistical approach to minimize multiextremal functions with nonconvex constraints in parameter identification problem of regional economic models developed at the Dorodnitsyn Computing Centre of the Russian Academy of Sciences [1]. In general, an identification problem of a mathematical model is to find its unknown parameters from optimal relations reflecting the degree of similarity between the calculated and actual (statistical) data. The optimization problem which arises here is time consuming and multiextremal. In the framework of the approach used, the solution of multi-dimensional problems is reduced to the solution of one-dimensional equivalents. The reduction is based on the use of Peano curves, uniquely mapping a unit interval on the real axis onto a hypercube. We also used a scheme to construct a set of Peano curves («rotating sweeps») which can be effectively applied to solve the problem on a cluster with dozens or hundreds of processors.
|
