Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
CHAPTER 7 L08: FREEVAL-RL 7.1 Overview of FREEVAL-RL The L08 product, FREEVAL-RL, is a much more complicated tool than either the C11 Reliability Analysis Tool or the L07 Analysis Tool. The FREEVAL-RL tool is based upon the FREEVAL model, which implements the freeway modeling methodologies founding the 2010 Highway Capacity Manual (HCM). The FREEVAL-RL version of FREEVAL enables the user to test the reliability impacts of projects by dynamically modeling multiple operating scenarios along the facility using a Monte Carloâtype strategy. Under this strategy, a scenario generator reruns a seed file multiple times to simulate different days along the facility. Since FREEVAL-RL is a more complicated tool, it has the potential to model reliability on a facility more accurately. The tool requires the user to designate and define several highway segments along the facility. This allows the tool to match the traffic impact in each segment and traffic queuing to spread from one segment to the rest. The study team found that this level of detail requires significantly more time to calibrate than the other SHRP 2 reliability tools. The calibration process is roughly comparable to that of a microsimulation model in terms of time and technical knowledge required. 7.2 Limitations of FREEVAL-RL Based on its testing of the FREEVAL-RL and communication with the tool developer, the study team found some limitations with the tool. These limitations are described in more detail below. In addition, the study team provides suggestions on how to address these limitations and improve the tool for implementation. Inefficiency of Data Inputs The FREEVAL-RL tool requires the user to enter various input data (including geometry data, segment type data, and demand flow data) cell by cell. This data entry method is slow and time- consuming. It is also easy for the user to make mistakes. Preferred methods are to allow the user to copy and paste data or allow the user to import the network geometry and demand flows stored in an Excel or text file. Segment Types and Study Period Time Cannot Be Changed Once the user creates a seed file, the study network and its study periods are configured and cannot be changed. This creates a number of limitations in the use of FREEVAL-RL for testing the reliability impacts of alternative strategies or projects: ⢠The user may want to increase or decrease the number of segments for an existing FREEVAL-RL model. This is particularly common as the seed file is generated and the 132
user tries to calibrate the seed file to baseline facility conditions. Currently, the user must re-input all data if the number of segments in the facility modeled is changed. ⢠The user may want to model both a.m. and p.m. periods, which may include different numbers of hours. If the user already has an a.m. model, it is tempting to base the p.m. model on the already coded a.m. model. However, the current version of FREEVAL-RL requires the user to create an entirely new model (i.e., the seed file) and re-enter all input data for the p.m. period. It would be more efficient to allow the user to edit the number of time intervals and the number of segments in the program. Such a change would support seed file calibration and alternatives testing. Maximum Number of Lanes FREEVAL-RL allows the user to define a mainline segment with one to six lanes in one direction. In addition, the maximum number of on-ramp and off-ramp lanes is limited to two lanes. Southern California freeways frequently have segments with more than six lanes in one direction as well as on-ramp or freeway-to-freeway connectors with more than two lanes. The study team understands that the FREEVAL-RL analytics are based on HCM 2010 and the tool shares its limitations, but extrapolating the methodologies and allowing the user to enter more lanes is essential for FREEVAL-RL to fit the real-world geometries found in Southern California. Through a detailed analysis of the FREEVAL-RL seed file results and trial and error, the study team found that the user can paste higher numbers (e.g., seven lanes for a freeway mainline segment and three for an on-ramp or off-ramp segment) in the file. FREEVAL-RL is able to use a higher capacity for a freeway mainline segment with more than six lanes. So it is possible to trick the model for these wider highway segments. However, the tool is not able to model the demands for a three-lane on-ramp despite the ability to paste three lanes in the highway segment definition. The study team found that a workaround is to model a short two-lane on-ramp segment followed by a second one-lane on- ramp segment. Given the prior limitation in changing the segment types, the study team had to re-enter the seed file data multiple times to find the appropriate workaround. FREEVAL-RL should be modified or the user guide (ITRE 2013) updated to document these model limitations or workarounds. Model Freeway Connectors FREEVAL-RL is not able to serve the demands from a three-lane freeway-to-freeway connector with high flow. The study team found that a workaround is to model a short two-lane on-ramp segment followed by another one-lane on-ramp segment. 133
No Network Geometry Viewer and Audit Tool It is easy to make mistakes when entering network geometries, particularly for large networks. FREEVAL-RL does not provide a graphical tool that can assist users in visualizing the results of segment coding. For example, the study team found that the wrong number of lanes had been entered for a highway segment after working on a model for several months. The team suggests developing a simple computer-aided design (CAD) map, such as the one found in FREQ, to show the number of lanes and other geometry data after the user inputs the geometry data. It will help the user develop models efficiently and help reviewers check the accuracy of models. In addition, the ability to associate highway postmiles with segment limits in FREEVAL- RL would help with auditing FREEVAL-RL coding and results. High-Occupancy Vehicle Lane or Managed Lane In California, high-occupancy vehicle lanes can be configured as either continuous-access or limited-access. The continuous-access configuration allows drivers to enter high-occupancy vehicle lanes at any time. The limited-access configuration limits high-occupancy vehicle ingress and egress to designated points spaced every few interchanges. Southern California freeways use the limited-access configuration exclusively, and nearly every freeway in Southern California has high-occupancy vehicle lanes. This means that to model improvements in Southern California, FREEVAL-RL must be able to handle the weaving and merging associated with limited-access high-occupancy vehicle lanes. In addition, PeMS data show that high-occupancy vehicle lanes typically have lower maximum throughput or capacity than other mainline freeway lanes do. According to the developer, FREEVAL-RL does not support any type of managed lane facility. As a result, one of the great challenges in using FREEVAL-RL for Southern California freeways is to find a way to model the limited-access high-occupancy vehicle lanes found in the region. The study team found an imperfect solution to modeling freeways with limited-access high-occupancy vehicle lanes after many months of trial and error. The solution is to ignore the travel conditions on the high-occupancy vehicle lanes and focus solely on the weaving and merging on the mainline freeway. The high-occupancy vehicle ingress/egress area is treated as a weaving segment with only the general-purpose (GP) lanes on the segment. Then, the weaving flows are estimated manually using rules of thumb (since these are rarely available from observed flows). The weaving from high-occupancy vehicle lanes to GP lanes is treated as an on- ramp flow, while the weaving from GP lanes to high-occupancy vehicle lanes is treated as an off-ramp flow. In addition, the traffic continuing on the high-occupancy vehicle lanes is treated as a ramp-to-ramp flow and set to a very low positive value. This ensures that the scenario generation does not result in negative flows. The study team arrived at this solution with the help of the FREEVAL-RL developers after running into problems with two other approaches to modeling high-occupancy vehicle lanes: 134
⢠The study team initially treated the high-occupancy vehicle ingress/egress segment as a standard weaving segment with a lane added to the adjacent GP lanes. This solution required estimating the weaving flow for the segment. Yet the scenario testing still failed with zero densities for some segments and unreasonable results for others. According to the developer, this was because the ramp-to-ramp flow (i.e., high-occupancy vehicle to high-occupancy vehicle flow for the high-occupancy vehicle ingress/egress segment) cannot be too high in FREEVAL-RL. However, for most high-occupancy vehicle ingress/egress segments on the test facilities (and for most freeways in the real world), the majority of the high-occupancy vehicle lane flow will continue on the high- occupancy vehicle lane rather than merging at the access point. ⢠The study team then coded the high-occupancy vehicle ingress/egress segment as a weaving segment with the same number of lanes as the GP lanes at the location with zero ramp-to-ramp flow. However, the team found that there was a slight chance that the scenario testing would lead to zero densities for some segments and unreasonable results in scenario runs. According to the developer, the team needed to avoid having zero ramp- to-ramp flow. This was fixed by assuming a slightly positive ramp-to-ramp flow. Even with the final solution adopted by the study team, the weaving segment function in FREEVAL-RL needs to be improved to deal with the various cases that may arise without crashing the program. In the longer term, the model would benefit from having an algorithm that handles limited-access high-occupancy vehicle lanes or managed lanes. Solving this problem will not only make the model more usable for Southern California, but also accommodate the growing interest in managed lanes across the nation. Weaving Segment and Volume Input When the study team ran a seed file, a weaving dialog box (called the âWeaving Volume Calculatorâ) would appear after clicking on the âRun the Seed File/Go to the Main Menuâ button. The team had to define the parameters for every weaving segment manually and adjust the weaving volumes for each interval. A more frustrating constraint was that the weaving segment inputs could not be saved for future use. This meant that reruns required recalculating and re-inputting the weaving volumes. Again, since most Southern California freeways have limited-access high-occupancy vehicle lanes, the weaving dialog box is critical to modeling Southern California facilities. The study team found this procedure to be very time-consuming and inefficient. If a weaving segment for a single time interval needed to be modified, the team had to re-enter all weaving data from the beginning. This procedure would be much more efficient if the weaving segment data could be entered, saved, and modified (perhaps through external spreadsheet files or in the initial seed file). Another innovation would be to allow the user to say âyes to allâ when allowing FREEVAL-RL to use the calculated weaving volumes (rather than forcing the user to review the weaving volume for every segment along the facility). 135
High Default Capacity Through extensive calibration testing, the study team found that the default capacity value in FREEVAL-RL was too high for the tested facility. This capacity can be modified through adjustments to the capacity adjustment factor (CAF) until the capacities calibrate to real-world traffic flows. The study team used PeMS data to perform this calibration. However, detection errors may result in a biased CAF estimate. As a result, the study team spent an extended effort on obtaining and processing sensor data from PeMS to estimate CAFs. The team further adjusted the estimated CAFs to achieve relatively reasonable results. Through this process, the team found the CAFs need to be adjusted for all segments. This suggests that the procedures used in FREEVAL-RL (and the HCM 2010) may overestimate capacity. Although the CAF procedure appears to be the correct fix, the study team found that after adjusting CAFs in the seed file, they were overwritten by the FREEVAL-RL engine and replaced with the default values (i.e., CAF = 1.0) during scenario testing. The study team has reported this problem to the developer, who has confirmed that it is a bug that will be fixed. Unexpected Results by Revising Result File There are two ways to generate a result file. The first is to run a seed file. The second is to modify an existing result file. Revising a result file would be a useful approach to implementing some of the workarounds described above. However, the study team found that the output from revising a result file was unpredictable and could produce different outputs than running the seed file, even if the inputs were the same. This is a potential programming error. To avoid this issue, the study team had to output results from the seed file. This was very time-consuming because the team had to go through the âWeaving Volume Calculatorâ procedure each time. Speed Contour Map The study team found the three-dimensional speed contour maps produced automatically in FREEVAL-RL hard to read. As a result, the team reviewed the values in the speed tab manually and generated speed contour maps externally using its own tools. The speed contour maps presented in this chapter were generated using these external tools. Ramp Merging Model The on-ramp flow for mainline segments in congestion may not be fully served because HCM 2010 gives the mainline flow a higher priority than the on-ramp flow. The study team found through its testing that the FREEVAL-RL model does not allow the vehicles on the ramp to merge to the freeway mainline if the mainline is congested and the on-ramp flow is high. As a result, not all of the on-ramp vehicles can be served, and the queues on the mainline were shorter than those found in real-world conditions. The ramp merging module (or the HCM 2010 algorithms) need to be adjusted to acknowledge the equal priority given to both mainline vehicles and merging vehicles. 136
Inability to Reflect Capacity Drop FREEVAL-RL does not have the ability to model capacity drop, which is a traffic flow characteristic observed from real-world point detector data. The study team found that the traffic flow performance is different when a queue is built up and when a queue is dissipated. Higher capacity is normally achieved when the queue is built up. Figure 7.1 shows a volume-occupancy plot based on data collected from a few days. Each data point is shown as a dot with different colors in the plot. The color theme is shown on the right. The figure clearly shows that at a certain occupancy level, the data points with higher volumes are collected in the early morning when the queue or congestion is forming. The data points with lower volumes are collected in the late morning when the queue or congestion is dissipating. FREEVAL-RL only allows users to provide a certain capacity value throughout the study period. Thus, the user may need to calibrate the model to have a relatively lower capacity in order to replicate the congestion period. Otherwise, the congestion will not remain as long as it should. Figure 7.1. Capacity drop. Documentation, Manuals, and Examples The study team reviewed the FREEVAL-RL documentation (ITRE 2013) extensively during its model testing. The team offers the following observations about the model documentation: 137
