Optimal Allocation of Testing Resources for Statistical Simulations (PREPRINT)

摘要

It is well known that statistical estimates from simulation, e.g., computation of mean and standard deviation, often involve significant uncertainty due to uncertainty in the input parameters. The uncertainty in the output results can be reduced with additional data on the inputs, e.g., additional experiments. An optimization methodology is proposed and implemented to determine the optimal additional experiments to conduct in order to minimize the variance in the output mean and standard deviation subject to a cost constraint. The number of additional experiments to add for each random variable depends upon several factors: the number of initial data points, the importance of the random variable in the response, the range of the random variable, and the cost of each experiment. The methodology is demonstrated using several numerical examples. The results indicate that the particle swarm optimization performs well and the solutions obtained are superior to ad hoc ones.