Statistical correlation in stratified sampling
Keywords: correlation, Monte Carlo simulation, Latin hypercube sampling, statistical analysis, sensitivity, reliability,
stochastic optimization, Simulated Annealing
ABSTRACT: A new efficient technique to impose the statistical correlation when using the Monte Carlo type
method for the statistical analysis of computational problems is proposed. The technique is based on the stochastic
optimization method called Simulated Annealing. The comparison with other techniques presently
used and intensive numerical testing showed the superiority and robustness of the method. No significant obstacles
have been found when also working with large problems (large number of random variables). The advantages
and limitations of the approach will be discussed. Remarks on the positive definiteness of target correlation
matrix are made. Numerical examples show the efficiency of the method.