A Comparative Evaluation of Two Global Search Algorithms

Two heuristic methods for locating the global optimum of a multimodal performance index surface are described. One method is based on a modified random creep procedure which first locates a local minimum and then searches the parameter space with vector steps whose mean length gradually increases. The second is a modification of the Kiefer-Wolfowitz stochastic approximation procedure, in which a random perturbation is added to each measurement. Both algorithms are compared by applying them to finding the roots of a nonlinear algebraic equation and to a constrained dynamic optimization problem.