ON AUTOMATED SEQUENTIAL STEADY-STATE SIMULATION

The credibility of the final results from stochastic simulation has had limited discussion in the simulation literature so far. However, it is important that the final results from any simulations be credible. To achieve this, validation, which determines whether the conceptual simulation model is an accurate representation of the system under study, has to be done carefully. Additionally, a proper statistical analysis of simulation output data, including a confidence interval or other assessment of statistical errors, has to be conducted before any valid inferences or conclusions about the performance of simulated dynamic systems, such as for example telecommunication networks, are made. There are many other issues, such as choice of a good pseudo-random number generator, elimination of initialisation bias in steady-state simulations, and consideration of autocorrelations in collected observations, which have to be appropriately addressed for the final results to be credible. However, many of these issues are not trivial, particularly for simulation users who may not be experts in these areas. As a consequence, a fully-automated simulation package, which can control all important aspects of stochastic simulation, is needed. This dissertation focuses on the following contributions to such a package for steady-state simulation: properties of confidence intervals (CIs) used in coverage analysis, heuristic rules for improving the coverage of the final CIs in practical applications, automated sequential analysis of mean values by the method of regener