On adaptive random search techniques

Several adaptive random search techniques for the solution of optimization problems involving noisy multimodal hills have been proposed in the literature. In this paper, the convergence properties of several of these methods are compared experimentally using digital simulations. In these methods, the performance index is evaluated at each iteration of the search. The rate at which the performance index, evaluated at the sample points, decreases as a function of the number of iterations is used as an estimate of the convergence rate of each method. The comparisons are based on average results obtained using several simulation runs. These runs start at random points. Therefore, the effects of the starting point tend to be averaged out.