We continue to develop information-statistical approach to minimization of multiextremal functions in the case of non-convex constraints. The proposed approach is called the index method of global optimization. The solution of multi-dimensional problems is reduced to their one-dimensional equivalents. This reduction is based on single-valued Peano curve mapping a hypercube onto the unit segment on the real axis. The scheme of constructing a set of Peano curves («rotated evolvents») is also used which can be effectively applied to the problem solution on a cluster with tens and hundreds of processors. Special attention is paid to the two-level parallel index algorithm for global optimization that takes into account the hierarchical structure of modern cluster systems.
|
