PGS-COM: A hybrid method for constrained non-smooth black-box optimization problems: Brief review, novel algorithm and comparative evaluation

The main available direct search methods for black-box optimization are reviewed.PGS-COM, a hybrid method for non-smooth black-box optimization, is proposed.PGS-COM performance is assessed and compared with that of 11 alternative methods.PGS-COM turns out to perform better than the other methods on most problems.Complex and population-based algorithms compare favorably among the other methods.

In the areas of chemical processes and energy systems, the relevance of black-box optimization problems is growing because they arise not only in the optimization of processes with modular/sequential simulation codes but also when decomposing complex optimization problems into bilevel programs. The objective function is typically discontinuous, non-differentiable, not defined in some points, noisy, and subject to linear and nonlinear relaxable and unrelaxable constraints. In this work, after briefly reviewing the main available direct-search methods applicable to this class of problems, we propose a new hybrid algorithm, referred to as PGS-COM, which combines the positive features of Constrained Particle Swarm, Generating Set Search, and Complex. The remarkable performance and reliability of PGS-COM are assessed and compared with those of eleven main alternative methods on twenty five test problems as well as two challenging process engineering applications related to the optimization of a heat recovery steam cycle and a styrene production process.