The use of latin hypercube sampling for the efficient estimation of confidence intervals

Conference · Sun Jul 01 04:00:00 EDT 2012

OSTI ID:22107753

Grabaskas, D.; Denning, R.; Aldemir, T. ^[1]; Nakayama, M. K. ^[2]

---

1. Ohio State Univ., 201 W 19th Ave, Columbus, OH 43210 (United States)
2. New Jersey Inst. of Technology, 218 Central Ave, Newark, NJ 07102 (United States)

Latin hypercube sampling (LHS) has long been used as a way of assuring adequate sampling of the tails of distributions in a Monte Carlo analysis and provided the framework for the uncertainty analysis performed in the NUREG-1150 risk assessment. However, this technique has not often been used in the performance of regulatory analyses due to the inability to establish confidence levels on the quantiles of the output distribution. Recent work has demonstrated a method that makes this possible. This method is compared to the procedure of crude Monte Carlo using order statistics, which is currently used to establish confidence levels. The results of several statistical examples demonstrate that the LHS confidence interval method can provide a more accurate and precise solution, but issues remain when applying the technique generally. (authors)

• Research Organization: American Nuclear Society, 555 North Kensington Avenue, La Grange Park, IL 60526 (United States)
• OSTI ID: 22107753
• Country of Publication: United States
• Language: English

Related Subjects

22 GENERAL STUDIES OF NUCLEAR REACTORS
DATA COVARIANCES
DATA PROCESSING
FINANCIAL DATA
MONTE CARLO METHOD
NUCLEAR POWER PLANTS
RISK ASSESSMENT
SAMPLING
STATISTICS