Search

The control variate method is a well-known variance reduction technique for discrete event simulation. This article explores the variance reduction achieved by employing an external analytical model to consolidate multiple input random variables into a single 'Analytical' Control Variate (ACV). Previous research suggests that this approach can produce significant variance reduction, but the resulting point estimate of the simulation response may exhibit unacceptable bias. A general Monte Carlo sampling method for resolving the bias problem is developed and demonstrated through a queueing network example. In order to use the method, the means, variances, and approximate distributions of the random variables used to produce the ACV must be known. In many cases, these requirements are not difficult to satisfy. With the Monte Carlo modification, the ACV method performs favorably when compared with classical internal and external control variate approaches. We compare the different control variate methods based on confidence interval width reduction, realized coverage, and estimated mean-square-error. We also compare the computational efficiency of the ACV method with that of uncontrolled simulation., 1. Introduction The method of control variates is an effective and practical Variance Reduction Technique (VRT) for discrete event simulation. VRTs are used to increase the efficiency of a simulation [...]

A study of airfield operations is presented with emphasis on the development of a model of the nested fork-join queueing network. Algorithms are used to examine the queueing network and the benefits of the model are described.

Planning and execution of military airlifts is largely based on simulation and network optimization models, where operations at individual airfields are usually represented by a few combined parameters. Historically, detailed airfield operations have been studied through simulations. A new analytical queuing model of military airlift operations at a single airfield is presented. The system is viewed as a capacitated open network of fork-join constructs and multiserver queues, and offers fairly accurate estimates of mean performance times.

The effect on the relative performance of series-system reliability estimates using very small binomial samples was investigated. A set of Bayes interval estimators was then compared with a set of classical estimators. Results reveal that, during Monte Carlo simulation, Bayes estimators elicited shorter interval estimates where the mean of prior system-reliability varied from about 20% or less of the true reliability.

Three analytical queueing models for an airport departure process have been developed and evaluated. The models represent the departure time as non-homogeneous Poisson process and the service time as suitable mixtures of exponential stages. The models can enhance departure capacity estimation and departure delay prediction at major airports. The performance of the Erlang-k model is simpler, more accurate and computationally efficient than the exponential model and the Erlang-k model with server absences.