Practical Simulation Budget Allocation for Ranked Subset Partitioning

Stochastic simulation is a powerful tool for analyzing complex discrete-event dynamic systems; however, it does not exhibit sufficient efficiency because of the requirement of numerous replicated simulations for obtaining accurate analysis results. Ranking and selection (R&S) efficiently allocates a simulation budget using ordinal optimization to correctly select alternatives of interest. Existing R&S methods focus on selecting an optimal alternative or a subset of optimal alternatives. Based on a generalization of this methodology, we propose an R&S method for partitioning
$k$
alternatives into
$n$
(
$2\le n\le k)$
exclusive ranked subsets, which is effective for job distribution and web search applications. The proposed method evaluates if the observed simulation results for each alternative have sufficient precision to correctly distinguish between the ranked subsets. It sequentially allocates a small portion of the budget based on the evaluation results, gradually improving the precision to maximize the efficiency. The superior efficiency of the proposed method compared with that of the existing methods is demonstrated using various numerical experiments. Furthermore, a practical problem that involves relocation-zone distribution in bicycle-sharing systems demonstrates that the proposed method can be effectively applied in situations requiring high simulation efficiencies, such as digital twins in complex systems.