State Transformations for Regenerative Sampling in Simulation Experiments

The randomness of the input variables in simulation experiments produce output responses which are also realizations of random variables. The random responses make necessary the use of statistical inferences to adequately describe the stochastic nature of the output. The analysis of the simulation output of non-terminating simulations is frequently complicated by the autocorrelation of the output data and the effect of the initial conditions that produces biased estimates. The regenerative method has been developed to deal with some of the problems created by the random nature of the simulation experiments. It provides a simple solution to some tactical problems and can produce valid statistical results. However, not all processes can he modeled using the regenerative method. Other processes modeled as regenerative may not return to a given demarcating state frequently enough to allow for adequate statistical analysis. This paper shows how the state transformation concept was successfully used in a queueing model and a job shop model. Although the first example can be analyzed using the regenerative method. it has the problem of too few recurrences under certain conditions. The second model has the problem of no recurrences. In both cases, the state transformation increase the frequency of the demarcating state. It was shown that time state transformations are regenerative and produce more cycles than the best typical discrete demarcating state in a given run length.