A research on simulation budget allocation and its application for optimizing the reliability of transportation system capacity

Discrete-event systems (DES) simulation is a popular tool for analyzing systems and evaluating decision problems since real situations rarely satisfy the assumptions of analytical models. Due to the randomness and uncertainty of the systems, the total simulation cost can be extremely expensive. The general purpose of this research is to find out efficient algorithms in simulation budget allocation and apply them to large-system analysis. Paper I presents a new approach that can further enhance the efficiency of ordinal optimization, which is called Optimal Computing Budget Allocation (OCBA). The approach determines a highly efficient number of simulation replications or samples and significantly reduces the total simulation cost. Paper II presents an effective simulation-based approach to selecting from a finite number of candidate systems, the one system whose mean performance measure is closest to that of a reference system. In the problems concerned, stochastic simulation is needed to analyze the performance measures for the reference system and all candidate systems. The approach utilizes the similar idea as OCBA in Paper I. It determines a highly efficient allocation of simulation replications for all candidate systems and significantly reduces the total simulation cost. The reliability of transportation network system capacity is a critical factor for evaluating the impact of improvements or damages in a transportation network. Paper III proposes three approaches that combine OCBA with three simulation approaches, Monte Carlo simulation procedure, Scenario-based algorithm and Dual-based algorithm, to optimize the system capacity reliability efficiently. These approaches can speed up the simulation dramatically over those three simulation approaches with Equal Computing Budget Allocation (ECBA). They are applied to an augmented network based on US double-stack freight transportation network and STRACNET to analyze the impact of link loss.