Footnote 1. The analyst should try to design the model to geographically and temporally encompass all significant congestion to ensure that the model is evaluating demands rather than capacity; however, the extent of the congestion in many urban areas and resource limitations may preclude 100 percent achievement of this goal. If this goal cannot be achieved 100 percent, then the analyst should attempt to encompass as much of the congestion as is feasible within the resource constraints and be prepared to post-process the model's results to compensate for the portion of congestion not included in the model. |-BACK-|
Footnote 2. Continuing improvements in data collection, computer technology, and software will eventually enable microsimulation models to be applied to larger problems. |-BACK-|
Footnote 3. Available at https://ops.fhwa.dot.gov/Travel/Traffic_Analysis_Tools/traffic_analysis_tools.htm. |-BACK-|
Footnote 4. The one exception to this statement is the recently developed freeway systems analysis methodology presented in Highway Capacity Manual 2000 (HCM 2000), which does explicitly treat oversaturated flow situations. |-BACK-|
Footnote 5. Additional discussions on software selection criteria can be found in Traffic Analysis Toolbox, Volume II: Decision Support Methodology for Selecting Traffic Analysis Tools, and in Shaw, J.W., and D.H. Nam, "Microsimulation, Freeway System Operational Assessment, and Project Selection in Southeastern Wisconsin: Expanding the Vision" (paper presented at the TRB Annual Meeting, Washington, D.C., 2002). |-BACK-|
Footnote 6. Some managers might devote about 50 percent of the budget to the tasks that lead up to and include coding of the simulation model, including data collection. Another 25 percent of the budget might go toward calibration. The remaining 25 percent might then go toward alternatives analysis and documentation. Others might divide the resources into one-third each for data collection and model coding, calibration, and alternatives analysis and documentation. |-BACK-|
Footnote 7. This example problem is part of a larger project involving metering of several miles of freeway. The study area in the example problem was selected to illustrate the concepts and procedures in the microsimulation guide. |-BACK-|
Footnote 8. Of course, if the analyst does not believe this to be true, then the study area should be expanded accordingly. |-BACK-|
Footnote 9. If existing or future congestion on either the freeway or arterial was expected to last more than 1 h, then the analysis period would be extended to encompass all of the existing and future congestion in the study area. |-BACK-|
Footnote 10. For this example problem, a microsimulation tool without rerouting capabilities was selected. It was supplemented with estimates of traffic diversions from a regional travel demand model. |-BACK-|
Footnote 11. The forecasting of future demands (turning volumes, O-D table) is discussed later under the Alternatives Analysis task. |-BACK-|
Footnote 12. Some microsimulation models may allow (or require) the analyst to input additional geometric data related to the grades, horizontal curvature, load limits, height limits, shoulders, onstreet parking, pavement condition, etc. |-BACK-|
Footnote 13. Some models may allow the inclusion of advanced traffic control features. Some models require the equivalent fixed-time input for traffic-actuated signals. Others can work with both fixed-time and actuated-controller settings. |-BACK-|
Footnote 14. Project constraints, traffic counter limitations, or other considerations (such as a long simulation period) may require that counts be aggregated to longer or shorter periods. |-BACK-|
Footnote 15. There are many references in the literature on the estimation of O-D volumes from traffic counts. Appendix A, chapter 29, HCM 2000 provides a couple of simple O-D estimation algorithms. Most travel demand software and some microsimulation software have O-D estimation modules available. |-BACK-|
Footnote 16. The software-supplied default vehicle mix, dimensions, and performance characteristics should be reviewed to ensure that they are representative of local vehicle fleet data, especially for simulation software developed outside the United States. |-BACK-|
Footnote 17. It is not reasonable to expect the simulation model to reproduce observed speeds, delays, and queues if the model is using traffic counts of demand for a different day or time period than when the system performance data were gathered. |-BACK-|
Footnote 18. HCM 2000. |-BACK-|
Footnote 19. There is no guidance on precisely what constitutes a "large" difference in counts. The analyst might consider investigating any differences of greater than 10 percent between upstream and downstream counts for locations where no known traffic sources or sinks (such as driveways) exist between the count locations. Larger differences are acceptable where driveways and parking lots could potentially explain the count differences. |-BACK-|
Footnote 20. There is no guidance on precisely how large differences can be between the HCM and field measurements before the field measurements may be suspect. The analyst might consider 25-percent differences to be within the normal range of accuracy of the HCM and take a second look at the calculations and field measurements if the differences are greater than that. |-BACK-|
Footnote 21. The selected software had been developed with defaults appropriate to the United States, and this was considered to be sufficiently accurate for this planning study of ramp metering. |-BACK-|
Footnote 22. The HCM was used to estimate surface-street saturation flow rates because none of the intersections were currently operating nor were they expected to operate within 90 percent of their capacity. Thus, an estimate is sufficiently accurate. Field measurements were not required. |-BACK-|
Footnote 23. The HCM was used to estimate freeway capacity because there is no freeway congestion under existing conditions that would enable field measurement of capacities. Another method would be to observe capacities on a similar type of facility in the area that operated under congested conditions. This method may be preferred if a comparable facility can be found. |-BACK-|
Footnote 24. Some software programs do not always use a link-node scheme, while others allow the analyst to code both directions of travel with a single link. The two-way links coded by the user are then represented internally (inside the software) as two one-way links. |-BACK-|
Footnote 25. The analyst may find it desirable to split links for performance reporting purposes. For example, it may be desirable to report density separately for the 460 m (1500-ft) influence area on the freeway near a ramp merge or diverge. However, the analyst should be cautious about the potential impact of split links on the ability of the software to accurately simulate vehicle behavior. |-BACK-|
Footnote 26. Some software programs provide analytical modules that assist the analyst in displaying and aggregating the results for specific groups of links. This feature reduces the necessity of adopting a node-numbering convention; however, a numbering convention can still result in a significant labor savings when reviewing text output or when importing text results into other software programs for analytical purposes. |-BACK-|
Footnote 27. The software-provided default values for driver behavior (aggressiveness, awareness, etc.) were used in this example problem. They were considered to be sufficiently accurate for this analysis. |-BACK-|
Footnote 28. Some of these techniques may not be available or necessary for some software programs. |-BACK-|
Footnote 29. Analysts should not expect classic macroscopic traffic-flow concepts to apply at the microscopic individual-vehicle level. Macroscopic flow concepts (e.g., no variance in mean speed at low flow rates) do not apply to the behavior of an individual vehicle over the length of the highway. An individual vehicle's speed may vary over the length of the highway and between vehicles, even at low flow rates. Macroscopic flow theory refers to the average speed of all vehicles being relatively constant at low flow rates, not individual vehicles. |-BACK-|
Footnote 30. A TMC with high-density camera spacing will be very helpful in reviewing the working model. Many TMCs are now providing workstations for traffic analysis/simulation staff. |-BACK-|
Footnote 31. For example, it could be that the warning sign for an upcoming off-ramp is posted in the real world 0.40 km (0.25 mi) before the off-ramp; however, because the model uses warning signs to identify where people start positioning themselves for the exit ramps, the analyst may have to code the warning sign at a different location (the location where field observations indicate that the majority of the drivers start positioning themselves for the off-ramp). |-BACK-|
Footnote 32. For example, a drawbridge that opens regularly might be coded as a traffic signal. |-BACK-|
Footnote 33. An example of this would be coding to address freeway drivers who do or do not make U-turns at an interchange (i.e., get off and then get back on the freeway in the opposite direction). |-BACK-|
Footnote 34. Bloomberg, L., M. Swenson, and B. Haldors, Comparison of Simulation Models and the Highway Capacity Manual, Preprint, Annual Meeting, TRB, Washington, DC, 2003. |-BACK-|
Footnote 35. The analyst determines the soundness of the simulation software when selecting it for use in the study. |-BACK-|
Footnote 36. Historically, it has been the practice to calibrate microsimulation models to all the traffic counts in the field. The majority of these counts will be at noncritical locations. The recommended strategy is to focus (at this point in the calibration process) only on the critical counts at the bottlenecks and to get the model to reproduce these counts correctly. Once this has been done, the rest of the counts are used later to check the route choice aspects of the model. All of this presupposes that the demands have been entered correctly and have already been checked against the counts at the entry gates as part of the first step (error checking). |-BACK-|
Footnote 37. It is certainly possible to use link-specific parameters exclusively during the capacity calibration step; however, this eliminates the benefits of the global parameter adjustment. The global adjustment ensures the accuracy of the model-predicted capacities on all links (even those not currently congested). Adjustment of link-specific parameters ensures the model accuracy only for the specific link. |-BACK-|
Footnote 38. The headways for the first three vehicles are discarded. |-BACK-|
Footnote 39. Microsimulation models assign driver-vehicle characteristics from statistical distributions using random numbers. The sequence of random numbers generated depends on the initial value of the random number (random number seed). Changing the random number seed produces a different sequence of random numbers, which, in turn, produces different values of driver-vehicle characteristics. |-BACK-|
Footnote 40. Some researchers have calibrated models using the percent MSE to avoid the unintended weighting effect when combining different measures of performance (such as volumes and travel time) into one measure of error. The percent MSE divides each squared error by the field-measured value. The effect of using percent MSE is to place greater weight on large percentage errors rather than on large numerical errors. The simple MSE is recommended for calibration because it is most sensitive to large volume errors. |-BACK-|
Footnote 41. Since the objective is to minimize error, dividing by a constant R (the number of repetitions) would have no effect on the results. However, R is included in the objective function to emphasize to the analyst the necessity of running the model several times with each parameter set. |-BACK-|
Footnote 42. The specific route choice algorithm parameters will vary by software program. They generally relate to the driver's awareness of, perception of, and sensitivity to travel time, delay, and the cost of alternate routes. |-BACK-|
Footnote 43. For a single-facility network, if there are still some remaining volume errors after the capacity calibration step, then the input demands should be checked for errors. |-BACK-|
Footnote 44. Some software programs allow selection of the algorithm and its associated parameters. |-BACK-|
Footnote 45. "Traffic Appraisal in Urban Areas, Highways Agency," Design Manual for Roads and Bridges: Volume 12, Section 2, Department for Transport (formerly Department of Environment, Transport, and the Regions), London, England, May 1996 (http://www.official-documents.co.uk/document/deps/ha/dmrb/index.htm). |-BACK-|
Footnote 46. Further discussions of Theil's Inequality Coefficient can be found in the Advanced CORSIM Training Manual, Minnesota DOT, 2002, and in Hourdakis, J., P. Michalopoulos, and J. Kottommannil, "Practical Procedure for Calibrating Microscopic Traffic Simulation Models," TRB, TRR 1852, Washington, D.C., 2003. |-BACK-|
Footnote 47. Actually, the analyst could (in theory) calibrate the percentage turns that were input at each intersection and ramp junction to better match the observed link flows downstream from the source nodes; however, this is currently impractical (there are too many variables to be adjusted without the assistance of a good computerized optimization algorithm). |-BACK-|
Footnote 48. A sketch-planning analysis using HCM techniques can be performed, if desired, to identify conceptual project alternatives prior to the development of a calibrated simulation model. The calibrated simulation model can then be used to refine the conceptual plans. |-BACK-|
Footnote 49. Deficiencies are identified by comparing the measured performance against the project objectives and agency performance standards identified in task 1 (Project Scope). |-BACK-|
Footnote 50. In support of the purpose and scope of the project, this identification of alternatives must take into consideration the range of demand conditions, weather conditions, incidents, and operational management strategies. Additional information and background may be found in the following documents:
Footnote 51. This report will be less useful if the analyst has split long sections of roadway into arbitrary short links for other reporting purposes. The result may be many "false alarms" of blocked links that do not actually obstruct an upstream intersection. |-BACK-|
Footnote 52. Many statistical phenomena approximate a normal distribution at large sample sizes. Even though most microsimulation analysts usually work with relatively few model repetitions, the assumption of normal distribution is usually good enough for most analyses. |-BACK-|
Footnote 53. Note that when computing the 95th percentile queue on the macroscopic level, it is typically assumed that the arrival of the vehicles are Poisson distributed. Microsimulation models predict the arrival patterns of vehicles, so the Poisson distribution assumption is not necessary when estimating 95th percentile queues using microsimulation data. |-BACK-|
Footnote 54. See example in Appendix B. |-BACK-|
Footnote 55. Those with a more sophisticated statistical aptitude may elect to use variance reduction techniques that employ a single common random number seed to reduce the number of required repetitions. These techniques are described in Joshi, S.S., and A.K. Rathi, "Statistical Analysis and Validation of Multi-Population Traffic Simulation Experiments," Transportation Research Record 1510, TRB, Washington, DC, 1995. |-BACK-|
Footnote 56. The analyst might consult Predicting Short-Term and Long-Term Air Quality Effects of Traffic-Flow Improvement Projects, Final Report, NCHRP Project 25-21, TRB, Washington, DC, 2003. |-BACK-|
Footnote 57. Some software programs also produce static graphs that can be very useful for gaining insight into the input or the results. |-BACK-|
Footnote 58. Animation output shows the results from just one run of the simulation model. Drawing conclusions about traffic system performance from reviewing just one animation result is like trying to decide if the dice are fair from just one roll. One needs to roll the dice several times, tabulate the results, and compute the mean and standard deviation of the results to have the information needed to determine if the dice are fair. |-BACK-|
Footnote 59. The analyst should verify with the software documentation or developer how statistics on blocked vehicles are accumulated in the travel time and delay summaries. |-BACK-|
Footnote 60. Person-miles traveled (PMT), if available, is the preferred measure of travel demand since it takes into account the number of people in each vehicle or mode. |-BACK-|
Footnote 61. Person-hours traveled (PHT), if available, provides a preferred measure of travel delay since it takes into account the number of people delayed in each vehicle. This is especially important for comparing the performance of HOV alternatives. |-BACK-|
Footnote 62. Note that the percentage confidence interval (such as a 95 percent confidence interval) has not been stated here, so it cannot be claimed that there is a certain probability of the true value falling within this 10 percent range. This is merely a sensitivity test of the impact of the demands being 10 percent lower or 10 percent higher than that forecast, without knowing the likelihood of it actually happening. |-BACK-|
Footnote 63. Note that density is NOT used as an LOS measurement for interrupted flow facilities, such as city streets with signals and intersections with stop signs. |-BACK-|
Footnote 64. Recurrent queues at signals that occur during each cycle are not considered to be building queues indicative of unserved demand that might bias the results for an alternative. |-BACK-|
Footnote 65. No increasing queues indicating underserved demand were found. |-BACK-|
Footnote 66. The improvement is modest enough that it might be worth establishing confidence intervals for the results and performing some sensitivity analysis and hypothesis tests to confirm the robustness of the conclusion. |-BACK-|
Footnote 67. With such a tight confidence interval, the analyst may be striving for a degree of precision not reflected under real-world conditions. |-BACK-|
Footnote 68. References: Hillier, F.S., and G.J. Lieberman, Introduction to Operations Research, Sixth Edition, McGraw-Hill, New York, 1995; Taha, H.A., Operations Research, An Introduction, Seventh Edition, Prentice-Hall, New York, 2003 (if the seventh edition is not available, look for the sixth edition published in 1996). |-BACK-|
Footnote 69. Adapted from Gardes, Y., A.D. May, J. Dahlgren, and A. Skabardonis, Bay Area Simulation and Ramp Metering Study, California PATH Research Report UCB-ITS-PRR-2002-6, University of California, Berkeley, February 2002. |-BACK-|
Footnote 70. If the analyst intends to perform hypothesis tests on only a few pairs of alternatives, then the equation provided should be sufficiently accurate. However, if the analyst plans to perform hypothesis testing of all possible pairs of alternatives, then this equation will underestimate the required number of repetitions needed to achieve the desired confidence level. The analyst should consult Lane, D.M., Hyperstat OnLine, An Introductory Statistics Book and Online Tutorial for Help in Statistics, Chapter 12: Introduction to Between-Subjects ANOVA ("All pairwise comparisons among means . . ."), Rice University (http://www.davidmlane.com/hyperstat). |-BACK-|
Footnote 71. Note that this is a two-sided t-test for the null hypothesis that the means are equal versus the hypothesis that they are different. |-BACK-|
Footnote 72. Analysts should consult standard statistical textbooks for tables on the type II errors associated with different confidence intervals and sample sizes. |-BACK-|
Footnote 73. Adapted from Lane, D.M., Hyperstat OnLine, An Introductory Statistics Book and Online Tutorial for Help in Statistics, Rice University (http://www.davidmlane.com/hyperstat). |-BACK-|