Extending Orthogonal and Nearly Orthogonal Latin Hypercube Designs for Computer Simulation and Experimentation
Abstract
Computational experimentation is an important tool of the military. It provides useful insights at a lower cost of time and money when compared to physical experiments. Consequently computational experiments are used to evaluate weapon systems for technology acquisibon examine tactics and to help select among alternatives for military operations and war plans. Experiments often consist of a large number of factors. Advancements in computing power and design of experiments (DOE) for simulation allow for the investigation of more of these factors through computational experiments achieved with less expense in time effort and money. Within the framework of DOE this thesis investigates Orthogonal Latin Hypercube (OLH) and Nearly Orthogonal Latin Hypercube (NOLH) designs. These designs are often used for computational experiments. This research greatly expands upon the size (in terms of runs and especially variables) of the available OLH and NOLH designs. Previously the largest catalogued OLH and NOLH designs were a maximum of 29 variables and 257 runs. OLH and good space-filling NOLH designs for up to 512 variables in 1025 runs are now available. This thesis also develops an algorithm for handling discrete factors with the designs. Finally the effects of stacking multiple OLH designs into one larger design are quantified. All of the designs developed in this research are available at the Simulation Experiments & Efficient Designs (SEED) Center website (htto://harvest.nos.edu).
Document Details
- Document Type
- Technical Report
- Publication Date
- Dec 01, 2006
- Accession Number
- ADA460403
Entities
People
- Ang Keng-ern Joshua
Organizations
- Naval Postgraduate School