Gornov A.Y.   Аникин А.С.  

Computing technologies for nonconvex parametric identification problem

Reporter: Gornov A.Y.

The problem of search of optimum parameters of the models described by the ordinary differential equations in a normal form of Cauchy is considered. As the quality criterion a wide class of non-linear functionals as integral and terminal type can be used. In the general formulation studied the problem can be interpreted as a problem of optimal control of a dynamic system in which the control actions are constants. To solve this problem a number of search algorithms is implemented aimed at optimizing non-convex functionals. A set of algorithms includes both traditional methods multistart, stratified sampling, Luus-Yaakola, Rastrigin and their modifications, and new optimization algorithms, focused on this class of problems: a spherical search methods, methods of dimension reduction, tunnel and other methods. The proposed algorithms are implemented in the software package OPTCON-A.
The report examines the computational optimization technology based on algorithms implemented, and the results of computational experiments on a collection of test and model problems.
This work was partially supported by RFBR grant number 12-01-00193 and SB RAS integration project number 83.