Test Problems for Parallel Algorithms of Constrained Global Optimization

This work considers the problem of building a class of test problems for global optimization algorithms. The authors present an approach to building multidimensional multiextremal problems, which can clearly demonstrate the nature of the best current approximation, regardless of the problems dimensionality. As part of this approach, the objective function and constraints arise in the process of solving an auxiliary approximation problem. The proposed generator allows the problem to be simplified or complicated, which results in changes to its dimensionality and changes in the feasible domain. The generator was tested by building and solving 100 problems using a parallel global optimization index algorithm. The algorithm’s results are presented using different numbers of computing cores, which clearly demonstrate its acceleration and non-redundancy.