Application of Deterministic Low-Discrepancy Sequences in Global Optimization

摘要: It has been recognized through theory and practice that uniformly distributed deterministic sequences provide more accurate results than purely random sequences. A quasi Monte Carlo (QMC) variant of a multi level single linkage (MLSL) algorithm for global optimization is compared with an original stochastic MLSL number test problems various complexities. An emphasis made on high dimensional problems. Two different low-discrepancy (LDS) are used their efficiency analysed. shown application LDS can significantly increase the MLSL. The dependence sample size required locating minima variables examined. found higher confidence in obtained solution possibly reduction computational time be achieved by total N. N should also increased as dimensionality grows. For clustering methods become inefficient. such multistart method computationally expedient.