A SCILAB FRAMEWORK TO TEST METAHEURISTICS: LAST RESULTS

Resumo

Many real-world optimization problems have been solved by metaheuristics in the last decades, mainly as a consequence of the increasing capacity of personal computers. In the same context, the number of available metaheuristics is huge. Furthermore, countless engineering problems are described as minimization/maximization procedures with adequate objective functions. In this scenario, the comparison between metaheuristics is frequently addressed in the specialized literature. In this work, we presented a modular framework (developed in Scilab language) devoted to the test of metaheuristics in several engineering problems (mainly in Chemical and Mechanical Engineering problems; as example we can cite high pressure phase equilibrium problems and inverse robot kinematics problem). The framework permits the selection of the metaheuristic itself as well as the control parameters (stopping criterion, number of individuals, specific parameters for each algorithm). Moreover, the computation structure presents a statistical comparison between the selected algorithms with respect to the number of function evaluations, iterations and computational time (including mean, standard deviation and a non-parametric statistical test), with minimal user-interference. New metaheuristics can be added to the framework using a standard coding procedure. The results are displayed in graphical form and can be exported using csv standard. Finally, we consider that this kind of computational structure is useful in the standardization of the procedures employed in the comparisons of these stochastic algorithms.