Consistent algorithm for sensitivity estimation of Markov discrete event dynamic system

Based on the characteristic of state transaction of Markov discrete event dynamic system (DEDS), efficient simulation algorithm for performance sensitivity estimation of such system is designed. The proposed algorithm can give consistent and reliable results for both steady state and transient sensitivity estimation, and has the advantage of easy implementation. The core formulae of our algorithm show that performance sensitivity of Markov discrete event dynamic system is the sum of the estimation obtained from both smoothed perturbation analysis (SPA) and likelihood ratios method (LR), therefore singly using any one of them usually cannot give consistent and reliable result. Finally, a two-server cyclic network is used as example to validate the proposed algorithm, both transient and steady state system performance sensitivities are estimated. The results are consistent and fit well with the theoretical values.