Flexible Selection for Efficient Latin Hypercube Design Optimization Method in Uncertainty Quantification

Abstract

Design of Experient is an integral part of uncertainty quantification. Latin hypercube designs, including methods such as maximizing distance LHDs and maximizing projection LHDs, are widely used for quantifying and computing engineering uncertainties. In particular, it is challenging to construct efficient designs for large designs with high parameter dimensionality and large data volume. In the current literature, various optimization tools exist to select the optimal LHDs, and each method has its advantages and disadvantages. In this paper, we combine commonly used optimization methods with different space-filling criteria to analyze the effects of parameter dimensionality, data volume, and algorithm hyperparameters for solving the problem of flexibility selection of optimization algorithms. The results of this paper can be used as a guide for the selection of optimization algorithms for experimental design aspects of practical engineering applications.