Agent based simulation output analysis

TL;DR: This work proposes a method that uses agent-based modeling to determine a truncation point to remove significant initialization bias and artificial bootstrap-like re-sampling of simulation runs is proposed for expensive simulations to estimate system performance sensitivity.

Abstract: In most realistic simulations there are multiple outputs of interest and the overall performance of the system can only be estimated in terms of these multiple outputs. We propose a method that uses agent-based modeling to determine a truncation point to remove significant initialization bias. Mapping the output of multiple replications into agent paths that traverse the sample space helps determine when a near steady state has been reached. By viewing these paths in reversed time, qualitative and quantitative methods can be used to determine when the multivariate output is leaving its near-steady state regime as the paths coalesce back towards their common initialization state. The methodology is more efficient and general than typical approaches for finding a truncation point for scalar outputs of individual replicates. Artificial bootstrap-like re-sampling of simulation runs is proposed for expensive simulations to estimate system performance sensitivity.