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.
Traffic Flow Data Preparation FREEVAL-RL requires users to provide traffic flow data at all cordon points in the modeled area. These cordon points include: (1) all on-ramps and off-ramps, and (2) the upstream end of the freeway mainline. For the high-occupancy vehicle ingress/egress weaving segment, the high- occupancy vehicle on-ramp flow and off-ramp flow were needed. FREEVAL-RL uses the flow data at the cordon points to calculate the demand for each segment and for each time interval. If the demand is higher than the capacity calculated based on HCM 2010, there will be congestion. Thus, if there is any data error for an on-ramp or off-ramp, the error may be amplified. For this reason, correct input flow data are very important to the analysis. A well-prepared I-5 CSMP dataset was utilized in this project. The dataset used different data sources, including Caltrans detector data from PeMS, turning movement data obtained from the Orange County Transportation Authority (OCTA), and data previously collected by data collection firms for the I-5 CSMP. The Caltrans detector dataset includes 29 days (i.e., Tuesdays, Wednesdays, and Thursdays from October 5, 2010, to December 16, 2010) and the median flow and speed values were used for model calibration. For the I-5 CSMP microsimulation model, the Caltrans detectors were the primary data source. However, the study team found that detectors were not present at some locations for the FREEVAL-RL testing. In addition, the detector data do not appear to be accurate for some locations. In these cases, the study team substituted data from another source. This was a time- consuming process to collect, compile, and analyze the data. Nevertheless, it was critical to get the correct data, because the input flow data are so important to FREEVAL-RL analysis. FREEVAL-RL models 15-minute time intervals and requires comparable input flow data. For the PeMS data, the study team only had 5-minute and hourly data previously generated for the I-5 CSMP. As a result, the study team had to improve its existing software tool to generate 15-minute data based on the 5-minute data directly from PeMS. The other data sources provided information in 15-minute intervals, so the raw data could be used directly. Estimates for the on- ramps (ingress) and off-ramps (egress) for the high-occupancy vehicle weaving segments came from high-occupancy vehicle detectors with good data quality upstream and downstream of the segments. 7.4 Model Calibration Initial Testing Figure 7.3 shows a speed contour map for the NB model in the a.m. period using observed data from PeMS. Figure 7.4 shows a comparable speed contour map from a FREEVAL-RL model calibrated using default parameters. While the real-world data show extensive congestion and queuing in the middle of the facility, no congestion is observed on the FREEVAL-RL speed contour map. The team decided not to use the speed contours provided by FREEVAL-RL, because the three-dimensional projection was hard to read. Figure 7.5 provides an example of a 142
FREEVAL-RL speed contour. The speed contour shown in Figure 7.4 was generated using a program developed by the study team to read the FREEVAL-RL speed output data and draw the speed contour based on a postmile-segment lookup table. Figure 7.3. Real-world speed contour map for the NB I-5 a.m. model. 143
Figure 7.4. Speed contour map from NB I-5 a.m. FREEVAL-RL model using default parameters. Figure 7.5. Example speed contour map from NB I-5 a.m. FREEVAL-RL model. 1 8 15 0.00 10.00 20.00 30.00 40.00 50.00 60.00 70.00 80.00 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51 Time Interval Sp ee d ( m i/h r) Segment Number Space Mean Speed Contours (mi/hr) 70.00-80.00 60.00-70.00 50.00-60.00 40.00-50.00 30.00-40.00 20.00-30.00 10.00-20.00 0.00-10.00 144
Model Refinement and Calibration To investigate the cause of the unmatched speed contours, the team checked the correctness of all cordon or demand flow input data, manually calculated the demands for each segment and time interval, and compared the calculated demands from FREEVAL-RL with the observed 15- minute flows from aggregated PeMS data for all segments. However, after fixing incorrect inputs found in on-ramp and off-ramp flow data, the team still could not replicate the real-world traffic congestion using FREEVAL-RL. After further examination of documentation in HCM 2010, the team found that FREEVAL-RL uses the highway capacity methodology, which overestimates capacity for this facility (which has four to five GP lanes and one to two high-occupancy vehicle lanes in the modeled area). Figure 7.6 shows the base capacities found in HCM 2010. By comparison, Figure 7.7 shows the speed-flow relationship for a mainline segment using 15-minute observed data collected from PeMS. As can be seen in the second figure, the maximum throughput for the facility is about 1,787 vehicles per hour per lane (vphpl). However, FREEVAL-RL estimates a capacity of about 2,293 vphpl for this segment. Free-Flow Speed (mi/h) Base Capacity (pc/h/in) 75 2,400 70 2,400 65 2,350 60 2,300 55 2,250 Figure 7.6. Base capacity from HCM 2010. Source: Highway Capacity Manual, 2010. Figure 7.7. Flow-speed relationship on I-5 based on 15-minute observed data. 145
The study team decided that it needed to calibrate the capacity in FREEVAL-RL to reflect the observed maximum throughputs in the PeMS data. FREEVAL-RL has the capacity adjustment factor (CAF), which can be used to adjust the capacity of a given segment. The CAF is equal to the actual capacity divided by the FREEVAL-RL capacity and by default is set to 1.0. By analyzing the 15-minute flow and speed relationship (as shown in Figure 7.7), the study team estimated the capacity by assuming that the maximum flow achieved is the actual capacity of the segment when congestion occurs. The study team was able to estimate the CAFs for segments having recurrent traffic congestion and detectors with good data (e.g., Segments 28 through 52 for the NB a.m. model). These CAFs were analyzed further to obtain a CAF pattern that associates a CAF value with the number of lanes and segment type, as shown in Figure 7.8. The CAF pattern was developed to help determine appropriate CAF values applied to segments (e.g., Segments 1-27 and 53-59 for the NB a.m. model) without traffic congestion in the base year. This CAF pattern is site specific and cannot be applied to other freeways or states that have different driver behaviors or roadway designs. Figure 7.8. CAF pattern on I-5. Model Calibration Results By applying the CAFs calculated above and further refining CAFs for some segments, FREEVAL-RL was able to produce much better results. For the NB a.m. model, the speed contour map from the finalized FREEVAL-RL seed file is as shown in Figure 7.9. Although it looks more reasonable, the traffic congestion occurs earlier and lasts longer than observed in the PeMS data (see Figure 7.3). 146
Figure 7.9. Speed contour map from the NB a.m. I-5 FREEVAL-RL model with CAF adjustment. For the SB p.m. model, the speed contour map from the finalized FREEVAL-RL seed file is shown in Figure 7.10. Compared with the real-world speed contour map shown in Figure 7.11, the calibrated FREEVAL-RL seed file results show less congestion. A detailed analysis indicated that the ramp merge model in FREEVAL-RL has the following limitations: ⢠FREEVAL-RL supports a maximum of two-lane on-ramps. The SB p.m. model has a major freeway-to-freeway connector at I-405. This connector has three lanes with more than 4,000 vphpl. A workaround suggested by the developer was to model the connector on-ramp as two continuous on-ramps. ⢠The merging model in FREEVAL-RL does not give priority to the traffic from the on- ramp, which causes the demands from on-ramps (especially around the I-405 interchange area) to be not fully served during congestion. As a result, the queue on the mainline is unrealistically short. 147
