Asymptotic Variance-based Sampling for Simulating Cycle Time - Throughput Curves

Simulation can be used to estimate a cycle time-throughput (CT-TH) curve of a manufacturing or service system model at several different traffic intensities. Cycle time variance increases rapidly as throughput approaches system capacity; thus, naïve sampling, which expends an equal amount of simulation effort at each traffic intensity being considered, leads to widely differing precision along the curve. In a manufacturing setting, the simulated CT-TH curve can aid in making startrate decisions so it is reasonable to assume that widely differing precision along the curve is an undesirable quality. The research presented here addresses this issue and suggests a fixed-budget variance-based sampling strategy that strives to achieve nearly equal absolute or relative precision. The proposed method determines the allocation of effort for each design point prior to any simulation effort being expended. Allocations are based on the asymptotic variance approximation at each design point relative to the asymptotic variance approximations at all design points investigated.