Collaborative Research: Effective Sequential Procedures for Risk and Error Estimation in Steady-state Simulation

点击翻译按钮获取中文摘要

This award provides funding for the development, implementation, and evaluation of a comprehensive methodology for steady-state simulation output analysis in a framework called Effective Sequential Procedures for Risk and Error Estimation in Steady-state Simulation (ESPRESS). Specifically, sequential methods for computing point estimators and confidence intervals for the steady-state mean and selected steady-state quantiles of an output process generated by a probabilistic simulation model will be developed. The confidence intervals will meet prespecified accuracy criteria and achieve a user-specified level of reliability in large samples. The steady-state mean describes long-run average performance, e.g., the average number of vehicles produced per month by an automotive assembly plant operating under a fixed capacity. Steady-state quantiles are used to measure the responsiveness of many service facilities and to characterize the risk associated with financial assets,e.g., the 90th percentile of call-waiting time is often used to evaluate the performance of a call center and the 5th percentile of the Value at Risk for a portfolio of financial assets is frequently used to describe the risk of loss on that portfolio. The implementation phase of ESPRESS will lead to computationally efficient statistical-estimation procedures that users can access and apply easily and quickly, which will be tested in the evaluation phase. The ultimate objective of this work is to provide practitioners and researchers with new techniques and public-domain software for the analysis of steady-state simulations that are completely automated, robust, and reliable as well as computationally and statistically efficient.If successful, the research will lead to fully automated sequential steady-state mean- and quantile-estimation procedures which are lacking in virtually all widely used commercial simulation packages. ESPRESS will be directly applicable to large-scale simulation studies in the governmental and military sectors as well as in a variety of industries, including aerospace, distribution, finance, healthcare, manufacturing, telecommunications, and transportation. Educational outreach will be a key component of this project, because all three investigators share a deep commitment to engineering education at both the undergraduate and graduate levels. Results of this project will be used in developing instructional modules for Project MINDSET (www.mindsetproject.org) and for new courses on Monte Carlo methods and their applications that will be developed at both the master's and doctoral levels at the two participating institutions.