Adaptive stochastic optimization using multiprocessors

摘要

A study is presented of a multiprocessor-based adaptive random search optimization algorithm that is suited for application to high-order systems with many parameters. Guidelines are established for optimum performance of the search technique in regard to the interaction of the optimization parameters and the influence of the number of processors being used. Analytical results are presented to show the influence of the random number distribution characteristics on the probability of successful improvement in the cost function, thereby suggesting more efficient search strategies. Furthermore, the relationship between the total number of processors and the corresponding rate of improvement in speed relative to sequential machines is presented. The class of problems likely to benefit from parallel processing is discussed. The comparative efficiency of the adaptive random search algorithm for different levels of parallelisms is illustrated through simulation studies.