Over the years, the meta-heuristics have been adapted and based on metaphors, whose proposals show effective solutions. Nevertheless, these abstractions are proving to be simple camouflage processes. This study presents a hyper-heuristic based on stochastic automata networks with learning, controlling a set of meta-heuristics. In this sense, this work investigates the moving through the search space performed by heuristic mechanisms, free of abstractions used by meta-heuristic. Or, in the conceptual terms proposed, we built up a hyper-heuristic model based on stochastic automata networks with learning, for selection and parameterization of low-level heuristics. The approach works on eight known meta-heuristic, exploiting features, and strengths of each algorithm. This identification allied to the theory of stochastic automata networks with learning guided the construction of their representations. These representations are consolidated in a meta-space, a part of the architecture of the hyper-heuristic proposed in work, named H2-SLAN Model. The results illustrate the effectiveness of our hyper-heuristic approach, regardless of heuristic composition, when compared to each meta-heuristic. Besides, hyper-heuristic performs better than individual meta-heuristics, thus significantly increasing optimization opportunities. As a result of this work, we got a system capable of selecting and parameterize low-level heuristics, with the ability to learn the heuristic movements employed by the model in the search space.