A Multi-Layer Line Search Method to Improve the Initialization of Optimization Algorithms
Résumé
We introduce a novel metaheuristic methodology to improve the initialization
of a given deterministic or stochastic optimization algorithm. Our objective
is to improve the performance of the considered algorithm, called
core optimization algorithm, by reducing its number of cost function evaluations,
by increasing its success rate and by boosting the precision of its
results. In our approach, the core optimization is considered as a suboptimization
problem for a multi-layer line search method. The approach
is presented and implemented for various particular core optimization algorithms:
Steepest Descent, Heavy-Ball, Genetic Algorithm, Differential Evolution
and Controlled Random Search. We validate our methodology by
considering a set of low and high dimensional benchmark problems (i.e.,
problems of dimension between 2 and 1000). The results are compared to
those obtained with the core optimization algorithms alone and with two
additional global optimization methods (Direct Tabu Search and Continuous
Greedy Randomized Adaptive Search). These latter also aim at improving
the initial condition for the core algorithms. The numerical results seem
to indicate that our approach improves the performances of the core optimization
algorithms and allows to generate algorithms more efficient than
the other optimization methods studied here. A Matlab optimization package
called ”Global Optimization Platform” (GOP), implementing the algorithms
presented here, has been developed and can be downloaded at:
http://www.mat.ucm.es/momat/software.htm
Domaines
Modélisation et simulation
Origine : Fichiers produits par l'(les) auteur(s)
Loading...