Variability Assessment for TRAF‐NETSIM

Abstract

The presence of high variability in traffic simulation results often leads to concern about their reliability, and consequently precludes a rigorous evaluation of the target traffic system's performance under various control strategies. Thus arises the need to assess the variability of key output parameters of interest, generally represented by the confidence interval. Since the operating characteristics of congested traffic flows yielded by a simulation program are a set of correlated sample observations, the conventional approach for computing confidence intervals via an approximate t‐distribution does not apply. This paper presents the statistical issues involved in the use of microscopic traffic simulation, followed by a critical discussion of variability assessment in congested traffic conditions. The batch‐means method, which allows the user to assess the variability of parameters, such as the average delay per vehicle, through a single relative long run, is introduced. The effectiveness of the batch means approach is evaluated via a set of simulation experiments wherein the output results of an example traffic system simulated by the TRAF‐NETSIM are examined. The research results, such as the batch size selection and the procedures for confidence interval computation, can either be built into a traffic simulation program as an automatic function or provided to traffic engineers as part of the users' guidelines.