On the Design of a New Stochastic Meta-Heuristic for Derivative-Free Optimization

Abstract Optimization problems are frequent in several fields, such as the different branches of Engineering. In some cases, objective function exposes mathematically exploitable properties to find exact solutions. However, when it is not case, heuristics appreciated. This situation occurs involves numerical simulations and sophisticated models reality. Then, population-based meta-heuristics, genetic algorithms, widely used because being independent function. Unfortunately, they have multiple parameters generally require numerous evaluations competitive solutions stably. An attractive alternative DIRECT, which handles a black box like previous meta-heuristics but almost parameter-free deterministic. its rectangle division behavior rigid, may many for degenerate cases. work presents an optimizer that combines lack stochasticity high exploration capabilities. method, called Tangram, defines self-adapted set rules search space yet relies on stochastic hill-climber perform local searches. expected be effective low-dimensional (less than 20 variables) few evaluations. According results achieved, Tangram outperforms Teaching-Learning-Based (TLBO), widespread plain multi-start configuration used.