Click for next page ( 67


The National Academies | 500 Fifth St. N.W. | Washington, D.C. 20001
Copyright © National Academy of Sciences. All rights reserved.
Terms of Use and Privacy Statement



Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.
Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.

OCR for page 66
APPENDIX A A SIMULATION-BASED APPROACH TO ATL EVALUATION These guidelines are primarily focused on deterministic approaches for predicting operations of an auxiliary through lane (ATL) adjacent to one or two continuous through lanes (CTL). The ATL volume prediction models described in Chapter 3 are based on a deterministic analysis framework and are directly compatible with the Highway Capacity Manual (HCM) procedures. The HCM recognizes that the use of alternative analysis tools, and specifically microsimulation approaches, has merit in a number of applications. In an effort to study the utility of microsimulation tools for ATL applications, this appendix describes guiding principles and considerations for applying simulation to ATL evaluation. The discussion covers both the operational evaluation of ATLs (delays, queue lengths, etc.), as well as approaches for estimating safety performance measures (conflicts) from simulation using the SSAM post-processing tool developed by FHWA (1). The lessons learned described in this appendix are closely related to the experience of the research team for NCHRP Project 3-98, which involved a significant microsimulation modeling and calibration effort. The project used the VISSIM simulation package (2), but this appendix describes the analysis principles in generic terms to the extent possible. Principles of Lane Change Algorithms Microsimulation tools explicitly model the movement of individual vehicles using a series of behavioral rules known as algorithms. Among these, lane changing algorithms are most critical for accurately describing ATL behavior. Most simulation models distinguish between "voluntary" and "mandatory" lane changes. Voluntary lane changes apply when a driver has multiple lanes available on the desired route, and switches lanes to--for example--pass a slower vehicle. The key point here is that the subject vehicle would have arrived at its desired destination regardless of whether it changed lanes. Mandatory lane changes on the other hand, are those that are necessary for performing a turning maneuver or for prepositioning in anticipation of a downstream lane drop. In other words, a mandatory lane change has to take place if a vehicle is to continue on its desired path. In the application to ATLs, the driver's decision to enter the ATL is generally a consequence of a voluntary lane change (e.g., to pass a queue of vehicles in the CTL). To be precise, the desire for a voluntary lane change is initially triggered by the car-following algorithm if the target vehicle's desired speed exceeds that of a vehicle ahead of it in the same lane. The voluntary lane change then describes the process of searching for suitable gaps in the adjacent lane (in this case, the ATL), and then ultimately switching lanes. On the other hand returning from the ATL to the CTL represents a mandatory lane change. Most simulation tools have different parameter sets in their voluntary and mandatory lane change algorithms and the analyst needs to understand the associated settings to accurately model the lane-changing behavior. Page A-1

OCR for page 66
In the case of mandatory lane changes, a common parameter used in the algorithm is the upstream decision distance. This distance is typically measured relative to the ATL lane drop and refers to the point at which drivers begin to be concerned with the lane drop. Exhibit A-1 illustrates this concept. Exhibit A-1 Illustration of Upstream Decision Distance in Simulation Direction of Flow Upstream Decision Distance The upstream decision distance describes the point at which the mandatory lane change algorithm becomes active. In most simulation tools, drivers will begin trying to merge at this decision point if gaps are available and will become increasingly aggressive about their lane-changing behavior as the distance to the downstream drop decreases. Further, in most cases, the mandatory lane change algorithm will override any voluntary lane changes. As a result, no voluntary lane changes will take place past the upstream ATL decision point, and consequently no CTL-to-ATL maneuvers will take place past that point. In this context, it is important to emphasize that a coded upstream decision distance that is greater than the total ATL length will prevent any voluntary lane changes into the ATL and will therefore result in zero through flow on the ATL. In NCHRP Project 3-98, this upstream decision distance (described in VISSIM as the lane change distance, LCD), proved to be the single best predictor of ATL lane utilization and a critical calibration factor to replicate field-observed ATL utilization in VISSIM. Calibration of Simulation Models Consistent with any simulation analysis, the parameter set used in the simulation model needs to be calibrated to match field conditio ns or known relationships in traffic flow theory. Calibration can include various changes to built-in simulation algorithms, including speed distributions, car-following logic, or lane-changing parameters. Significant research is available on the topic of simulation calibration, including material compiled by FHWA in the Traffic Analysis Toolbox. For this discussion, the topic of calibration is condensed to the specific application to ATLs. The foremost goal in the calibration of a simulated CTL-ATL system is to match the field-observed ATL utilization or, in the absence of field data, the ATL volume predicted from the models presented in these guidelines. By varying the LCD parameter in VISSIM, the research team was able to successfully calibrate 19 of the 22 studied ATL approaches. The remaining three approaches exhibited very low utilization (less than 10 percent). These low-utilization percentages could not be replicated without also making significant adjustments to the car- following logic, which in turn resulted in more simulated "crashes." Page A-2

OCR for page 66
Exhibit A-2 shows the resulting relationship between field-observed and simulated ATL percentages for the 19 approaches (black) and a best-fit line. The three low-utilization approaches (gray) are treated as outliers. Exhibit A-2 Calibration Result Showing Field- 50 Observed vs. Simulated ATL Utilization 45 Calibrated Sites R2 = 0.9499 Simulated ATL Utilization (%) 40 Low-Utilization Sites 35 30 25 20 15 10 5 0 0 5 10 15 20 25 30 35 40 45 50 Field-Observed ATL Utilization (%) The simulated ATL utilization percentages shown in Exhibit A-2 were the result of free lane selection by drivers on the intersection approach, subject to the algorithms of car-following, lane changing, etc. The ATL utilizations were not "forced" in the sense that a fixed percentage of through traffic was routed through the ATL. In this sense, the resulting R2 of 0.95 shows a high rate of success in calibrating ATL utilization through the LCD parameter in VISSIM. Other calibration efforts may include accurate coding of turning-movement flows, speed distributions, signal-timing parameters, etc. The analyst should further validate some of the outputs from the simulation model to field data if available. These outputs may include approach delays, total through travel time, or vehicle queues. ATL Utilization Prediction Model Given the sensitivity of the LCD parameter on ATL utilization, an effort was made to predict the correct LCD setting for (future) ATL sites, where the true utilization is unknown. The dependent variable LCD was expressed in the following way: LCD % Total: the LCD expressed as a percentage of the total ATL length, computed as Page A-3

OCR for page 66
Other forms of the dependent variable were explored, but the one quoted above emerged as the preferred definition. Several explanatory variables were hypothesized to affect the LCD used to calibrate VISSIM, including traffic volumes, approach speeds, upstream and downstream length, and a distinction between single and dual CTLs. Ultimately, the following two explanatory variables were used in the LCD prediction model: Volume: through traffic flow rate expressed in vehicles per hour (vph) Upstream: the length of the ATL segment upstream of the stop bar, in feet The resulting model predicting LCD%TOTAL as a function of these two variables is given below: R2 = 0.622 The R2 value suggests that 62.2 percent of the variability in the LCD variable that provided the best match to the field data is explained by the variables in the model for the regression data set. This suggests that the model can be used to arrive at a reasonable initial estimate for the LCD parameter if VISSIM is used to model the ATL. For other simulation tools, this model may similarly guide an initial parameter estimate, but the model has not been calibrated for such applications. The model suggests that the LCD begins at 89.696 percent of the total ATL length, which coincides approximately with the highest LCD value observed in previous VISSIM calibration. That term is then discounted with increasing upstream length and through volume. This implies that ATL utilization increases with increasing upstream length and through volume. This relationship is consistent with field observation. A closer exploration of the two explanatory variables also suggests a good model fit as shown in Exhibit A-3. The exhibit shows that LCD%Total is approximately linear with respect to the upstream length of the ATL and the combined through volume. Page A-4

OCR for page 66
Exhibit A-3 LCD%TOTAL VS. Upstream Length LCD%TOTAL VS. Through Volume Sensitivity of LCD%TOTAL vs. 100 100 Upstream Length and Through 90 Volume 90 80 80 70 70 LCD%Total LCD%Total 60 60 50 50 R2 = 0.5138 40 40 30 30 R2 = 0.5677 20 20 10 10 0 0 0 500 1000 1500 2000 0 700 1400 2100 Upstream Length (ft) Through Volume (vph) Surrogate Safety Assessment As part of the safety analysis, the researchers investigated a technique to use the Surrogate Safety Assessment Methodology (SSAM) in conjunction with a calibrated VISSIM model in order to predict safety conflicts at ATL approaches. SSAM was developed by FHWA as a post-processing tool to estimate vehicular conflicts from simulation trajectory files. These raw simulation output files store the speed, position, and acceleration of every simulated vehicle during each simulation time step among other data. From the trajectory files, SSAM defines a conflict, for example, if two vehicles occupied the same space within a user- defined time-to-crash (TTC) threshold. Full details on the SSAM tool and definitions of terms can be found in the documentation for SSAM (1). In this research, the SSAM tool was applied to all 16 of the ATL approaches that were part of the safety evaluation in Chapter 4. This evaluation was performed after the ATL utilization was calibrated to empirical observations. The results of this investigative conflict study were not fully validated by the crash data presented in Chapter 4, mostly due to the low crash sample sizes, flaws in crash reporting, and other errors. Nonetheless, the SSAM output can still be used to examine relationships between conflict frequency and key ATL design elements such as downstream length. From the calibrated simulation models, SSAM uses the trajectory (*.trj) files generated by the simulation and can apply various filters during analysis to define a conflict: Conflict type. SSAM distinguishes between angle, lane change, and rear-end conflicts by the angle at which the conflict occurs. Only lane change and rear-end conflicts were targeted in this analysis. Time to crash (TTC). The threshold for what defines a "conflict" is the TTC, which can be adjusted to 0.5, 1.0, or 1.5 seconds. This analysis used a TTC equal to 1.5 seconds to get a conservative estimate of the number of conflicts to compare with crash data. Link. The analyst can filter conflicts by the link ID number used in the simulation tool. This allows the analyst to filter conflicts by the upstream or downstream portion of the ATL (or by multiple ATL approaches within one Page A-5

OCR for page 66
simulated intersection) as long as they are numbered as separate links in the simulation network. Trajectory file. If multiple runs are simulated, the analyst can obtain the conflict frequency for each trajectory file within the multi-run simulation. Using these methods, the analyst can compare design alternatives such as downstream length, speed, congestion, and the presence or absence of an ATL. Exhibits A-4 and A-5 show how the number of SSAM rear-end and sideswipe conflicts, respectively, changed for an exclusive lane with respect to downstream ATL length and XT. Rear-end conflicts remained relatively consistent as downstream length increased, but the number of sideswipe conflicts spiked at a downstream length of 800 feet. This may be due to some quirk in the simulation or SSAM logic and the low sample size of sideswipe conflicts. Conflicts tended to increase fairly steadily with increasing XT, as might be expected. Exhibit A-4 250 SSAM Rear-End Conflict Comparison (No Right Turns) 200 150 Conflicts XT = 0.75 100 XT = 1.00 XT = 1.25 50 0 0 200 400 600 800 1000 1200 Downstream Length (ft) Exhibit A-5 16 SSAM Sideswipe Conflict Comparison (No Right Turns) 14 12 10 Conflicts 8 XT = 0.75 6 XT = 1.00 4 XT = 1.25 2 0 0 200 400 600 800 1000 1200 Downstream Length (ft) Page A-6

OCR for page 66
350 Exhibit A-6 SSAM Rear End Conflict 300 Comparison (200 Right-Turning Vehicles per Hour) 250 Conflicts 200 XT = 0.75 150 XT = 1.00 100 XT = 1.25 50 0 0 200 400 600 800 1000 1200 Downstream Length (ft) Exhibits A-6 and A-7 show the same types of comparisons for a shared ATL with 200 right turns per hour. The exhibits indicate that low-to-moderately congested approaches had low levels of conflicts when compared to those simulated at XT = 1.25. While rear-end conflicts remained relatively unaffected by changes in downstream length, the number of sideswipe conflicts tended to increase as downstream length increased. This could be explained by the exposure, as a greater downstream length tended to generate more conflicts in SSAM simply because the conflict area was lengthened . Note that if the number of conflicts was normalized by downstream length, a decreasing trend would emerge. Also note that there were many more conflicts generated by the shared lane than by the exclusive ATL scenario. Exhibit A-7 SSAM Sideswipe Conflict 70 Comparison (200 Right-Turning Vehicles per Hour) 60 50 Conflicts 40 XT = 0.75 30 XT = 1.00 20 XT = 1.25 10 0 0 200 400 600 800 1000 1200 Downstream Length (ft) Page A-7

OCR for page 66
The analysis of variance for the rear-end conflict data across all levels tested indicated that there were statistically significant interactions between the right- turn volume and XT and between the right-turn volume and the downstream length. The former interaction is intuitive, as the effect of right-turning vehicles as a detriment to safety magnifies when there is more through traffic in the shared ATL. In terms of the sideswipe conflicts, multiple significant interactions existed--some of these were unexpected and may be due to the low sample size for sideswipe conflicts. In general, it appears that the SSAM logic confirms intuition. First, it appears that changes in downstream length are not associated with changes in the observed number of conflicts in the ATL. Second, shared ATLs (at least those with 200 or more right-turning vehicles per hour) tend to have more conflicts than exclusive ATLs. Finally, the crash increase with the increase in XT discussed in Chapter 4 was supported by the increase in conflicts with XT shown in Exhibits A-4 through A-7, particularly as XT increased beyond 1.0. Proposed Work Flow of ATL Simulation Study If an analyst is studying the feasibility of an ATL intersection improvement, the following list of steps represent a proposed analysis work flow. Please note that additional steps may be necessary, depending on the specific location, and the practitioner should exercise sound judgment in any simulation analysis. Step 1: Gather input data, including existing and proposed intersection geometry, traffic turning movements (current and forecast), approach speed limits, and signal timing data. Step 2: Model baseline, representing the existing intersection without ATL approaches. Step 3: Calibrate baseline, by comparing the modeled operations to field data or other analysis approaches. Make any necessary adjustment to traffic volumes, speed inputs, signal timing, or other simulation algorithms. Step 4a: Estimate initial LCD parameter, using the predictive model in this appendix as a function of the total (future) through traffic flow and proposed total ATL length. An initial estimate of the ATL upstream and downstream lengths is therefore needed for this analysis. Step 4b: Estimate ATL predicted volume, using the models described in Chapter 3 of these guidelines. These estimates will be used to validate that the ATL utilization is modeled correctly. Step 5: Model ATL geometry, using proposed geometry, signal timing, and volumes (step 1), and the initial LCD parameter from Step 4a. Step 6: Calibrate ATL operations, by modifying the LCD until the simulated ATL volume matches (approximately) the predicted volume from Step 4b. As a general guidance, a longer LCD will result in lower utilization of the ATL. Step 7: Evaluate ATL performance, by running repeated iterations of the baseline and ATL scenarios and comparing the average performance. A suggested performance measure is the total through travel time, which is readily compared to field data. Additionally, approach delay and queue lengths are Page A-8

OCR for page 66
important performance measures that are also predicted in the deterministic HCM analysis approach. Step 8: Evaluate safety performance, by extracting trajectory files from simulation and analyzing for conflicts in the SSAM post-processing tool. To estimate ATL safety performance, the SSAM evaluation should be limited to the specific ATL link in question and should distinguish between rear-end and lane- changing conflicts, as well as conflicts in the upstream and downstream portions of the ATL. Depending on the objective of the analysis, it may be useful to obtain performance measures on a per-lane basis, to be able to isolate the performance of the ATL. As general guidance, it is important to use the same definitions of performance measures for baseline and any ATL scenarios to assure an even comparison. For example, any travel time segments should be defined for a distance long enough to contain the longest queue length in the baseline scenario and should not be changed when moving to the ATL scenario. References 1. Surrogate Safety Assessment Model (SSAM). Version 2.0. Federal Highway Administration. Siemens ITS: 2004. 2. VISSIM. Version 5.30-02. PTV America: Portland, Oregon, 2010. Page A-9