在对遍历Markov链的性能灵敏度分析中,可以用计算Markov链的实现因子来代替计算Markov链的性能势。给出了一种基于耦合技术的仿真方法来估计实现因子。使用经过优化的耦合矩阵,通过仿真来快速获得实现因子的估计。这种方法有两个优点:一是不用去求稳态分布,二是能减少仿真中的计算时间。
Realization factors can be computed instead of performance potentials in performance sensitivity analysis of an ergodic Markov chain. A method was given to estimate the realization factors by simulation based on the coupling approach. An optimized coupled matrix was used to obtain quickly the estimation for realization factors in simulation. In this way, explicit knowledge of the stationary distribution is not required and computation in simulation can be saved.