A simulation technique for estimation in perturbed stochastic activity networks

Abstract

We are concerned with what-if analysis in estimating the expected value and the distribution function of completion time in stochastic activity networks. Widely used Monte Carlo simulation models of stochastic networks are often subject to errors caused by the estitnated parameter(s) of underlying input distribution function. " What-if' analysis is needed to establish confidence with respect to small changes in the parameters of the input distributions. However, traditional "what-if'analysis requires a separate simulation run for each input value. Recently, a method based on Likelihood Ratio (LR) for estimating performance function far several scenarios using a single-run extrapolation has been presented. In this paper, we experiment the use of this LR method in a network with exponential arc durations. We also consider the method with a nonlinear control random variate (NCRV) and compare it to crude Monte Carlo. The results show that the NCRV method induces variance reduction and is an effeetive filter to stabilize statisti cal variation of this single-run estimate.