随机复杂系统设计方案排序问题的动态计量预算分配策略

Dynamic computing budget allocation for design ranking in stochastic complex systems

仿真技术已被广泛地用于评估与优化随机复杂系统,例如生产制造、网络通信、医疗健康等信息物理融合系统.然而,将仿真优化方法用于方案排序问题的研究甚少.本文考虑只能通过系统仿真才能获取性能表现的k个备选方案排序问题,在贝叶斯(Bayes)框架下,构建以寻求最大化后验方案正确排序概率为目标的动态仿真预算优化分配模型.通过将仿真预算分配决策过程表述成一个随机动态规划问题,推导出一种动态计量预算分配策略,并证明了该分配策略的渐进最优性.数值实验以及案例分析的结果表明,本文提出的算法显著地提高了方案排序问题的仿真优化效率.

Simulation has been widely used to evaluate and optimize stochastic complex systems such as manufacturing,telecommunication,and healthcare systems,which are known as cyber-physical systems.However,research on design ranking in simulation optimization is limited.This research considers the complete ranking of k designs,whose performance can only be evaluated through simulation.Using the Bayesian framework,an optimal dynamic computing budget allocation model is developed,which maximizes the posterior probability of correct rankings.The computing budget allocation model is reformulated as a stochastic dynamic programming problem.This paper derives the optimal dynamic computing budget allocation rule and confirms its asymptotic optimal property.Numerical experiments and a case study demonstrate that the proposed simulation procedure considerably increased the simulation optimization efficiency for design ranking problems.