The use of micro simulation tools to analyze and/or predict the performance of road sub-networks is now common practice among traffic engineers. Popular micro simulation tools, such as Paramics, Vissim, Integration, Aimsum, and NetSim are considered stochastic models in that they use pseudo random numbers to control random processes within the simulation, such as lane changing decisions, desired speeds, etc. As a result, traffic engineers and simulation model users typically carry out several model runs for each set of traffic and control conditions, each run with a different random number seed, in order to imitate the randomness in field observations. Often, the results from the replications are averaged in the hope that the mean of the multiple runs is a reliable predictor of the average conditions that would occur in the field. This paper presents the finding of a study that has been carried out to determine the extent to which the use of multiple runs each with a different random number seed captures the degree of variability typically present in real networks. Traffic counts from permanent count stations located on arterial roads in three North American cities were used to quantify the variability in peak hour traffic volumes for non-holiday week days. The results of this analysis suggest that these peak hour volumes are Normally distributed with a coefficient of variation of approximately 8.4%. In contrast, the use of random number seeds with the Vissim simulation tool produced a coefficient of variation of the peak hour volume of only approximately 3.5%, substantially less than that observed in the field data. The impact in terms of intersection performance (i.e. delay) was investigated. Two methods to estimate intersection performance were developed. Method 1 consisted of defining an average peak hour volume (for a given volume to capacity ratio) and introducing variability through the use of different random number seeds. This is the method typically used in practice. Method 2 consisted of using only a single random number seed but generating variability in peak hour traffic volumes by defining as input to the model peak hour volumes generated using a distribution based on field data. For each method, eleven traffic demand scenarios were developed encompassing intersection volume to capacity (v/c) ratios ranging from 0.6 to 1.10. Each scenario was replicated 100 times. The results show that: (1)Average intersection delay obtained via Method 1 is a biased estimated of the true mean delay. (2) The average intersection delay obtained via Method 1 may under-estimate the true mean delay by as much as 13%. (3) The coefficient of variation of intersection peak hour delay resulting from the use of Method 1 is approximately half of that expected. (4) The use of Method 1 tends to predict many fewer failures (i.e. peak hour intersection delay exceeding some threshold) than would actually be experienced.
Samenvatting