Value of Travel Time Reliability in Transportation Decision Making: Proof of Concept—Portland, Oregon, Metro (2014)

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.

29 C h a p t e r 5 5.1 Metro Dynust Dynamic traffic assignment Model establishment and Calibration For this project, Metro staff created a DynusT regional DTA model. Significant effort went into converting, coding, and debugging the regional model. Several challenging tasks, including setting up the signal timing and intersection turn bays, were carefully carried out by the Metro staff and sup- porting University of Arizona staff. Once the base model was set up, calibration was performed to ensure that the estab- lished DynusT model was consistent with the existing TDM from which prior scenario analysis results were produced. After several rounds of traffic flow model calibration, the final model parameters were chosen for freeways: • Alpha = 2.3 • Jam density = 190 • Minimum speed = 5 mph In addition, the freeway bias was set to 10%, and the entire regional model was run with all signals set to actuated (45-second maximum, 10-second minimum, 4-second amber) signals. The DynusT model was validated against INRIX and count data. All the calibration efforts resulted in improved overall travel times and volumes when compared against INRIX and count data. When plotted against the static assignment results from the existing TDM based on Emme, the PM two-hour vol- umes from DynusT in the Southwest Corridor were still low on the major arterial (Barbur Boulevard), with differences averag- ing 42% lower in DynusT than in Emme (Figure 5.1). There was a much better fit on the freeways, with an average difference of 4% and R2 = 0.9407. Based on these results, the signals were changed to known timings (pretimed) along Barbur Boulevard and the on- and off-ramps for I-5 in the Southwest Corridor to see if a closer fit could be achieved between the models. These changes resulted in much better fit in the study area, with arterials averaging just 12% difference and freeways still averaging 4% difference. The overall R2 was further improved to 0.9462 (Figure 5.2). Next, travel times were compared between the Emme model and the DynusT model. Table 5.1 shows AM peak- period two-hour and midday one-hour travel time compari- sons between the static model and DynusT. The statistics are categorized into three groups: all zones, either origin (O) or destination (D) in the Southwest Corridor, and both O and D in the Southwest Corridor. Examined from the standpoint of percentage difference (% Difference in Table 5.1) for all three groups, midday travel times looked satisfactory, with zonal travel time differences ranging from only 7.5% to -11.1%. However, the difference increased for the peak-period results. When comparing all zone pairs in the region (4,600,000+), there was still a signifi- cant difference (~32%) between the DynusT and Emme model travel times. Much of this difference can be attributed to the actuated signals, which reach a maximum at 45 seconds in DynusT. On many corridors in the region, however, the maxi- mum green time can exceed 100 seconds, and the edits to Barbur Boulevard in the model showed that changes in maxi- mum green time can make a big difference in route selection and zone-to-zone travel time. So, either the maximum green time allowed for select actuated signals should be increased, or pretimed signal plans along select corridors should be manu- ally input to get more reasonable regional DynusT travel times. When comparing all O-D pairs either beginning or ending within the Southwest Corridor (1,000,000+), there was an even bigger discrepancy between weighted travel times (~37%). However, when focusing on those zone pairs contained wholly within the Southwest Corridor study area (66,500+), there was a much better fit between the models (~20% difference, representing less than 2 min). Figure 5.3 shows the entire Portland regional highway net- work. The Southwest Corridor study area, shown in Figure 5.4, Analysis Model Preparation

30 Emme volumes 4 pm – 6 pm D yn u sT v o lu m es 4 p m – 6 p m Figure 5.1. Comparison of 4 p.m. to 6 p.m. volumes between DynusT and Emme. D yn u sT v o lu m es 4 p m – 6 p m Emme volumes 4 pm – 6 pm Figure 5.2. Improved 4 p.m. to 6 p.m. volume comparison between DynusT and Emme.

31 Table 5.1. AM Peak-Period Travel Time Comparison Between DynusT and Emme Zone No. of Zone Pairs Weighted Mean AM 2-hour SOV Travel Time Weighted Mean Midday 1-hour SOV Travel Time Emme (min) DynusT (min) Difference (Emme – DynusT) % Difference (from Emme) Emme (min) DynusT (min) Difference (Emme – DynusT) % Difference (from Emme) All zones 4,674,244 15.54 20.56 -5.02 -32.3% 13.18 14.64 -1.46 -11.1% O or D in SW Corridor 1,049,028 17.03 23.40 -6.37 -37.4% 13.96 14.87 -0.91 -6.5% O and D in SW Corridor 66,564 8.55 10.28 -1.73 -20.2% 7.20 6.66 0.54 7.5% Note: SOV = single-occupant vehicle. Figure 5.3. Portland regional highway network.

32 Figure 5.4. Southwest Corridor study area shown in coral. encompasses 258 traffic analysis zones; this area actually con- tains most of the high-employment and residential areas that were of concern for the purposes of this study. This area includes the Portland central business district (CBD), the Lloyd District, Washington Square, Lake Oswego, Tualatin, Tigard, and much of the industrial northwest area of Portland. The team decided to focus solely on DynusT and FAST-TrIPs travel times and reliability travel time equivalents (TTEs) only for those zone pairs contained within the study area and substi- tute Emme automobile and transit travel times for all other areas in the region. Using these times greatly improved the model calibration and allowed the team to move forward with TDM integration. The next step was the integration of DynusT and FAST-TrIPs and final integration of the dynamic models with the Metro TDM. 5.2 FaSt-trIps Model preparation and Coding 5.2.1 FAST-TrIPs Model Updates As part of this project, the L35A team continued to make sub- stantial improvements to the FAST-TrIPs model. The path search model (i.e., trip-based shortest path) in FAST-TrIPs was improved to be multithreaded, with significant speed-ups. In a test on a four-core machine, the run time was improved by up to 70%. A new submodel was added for passenger appearance at boarding stops. The submodel determines, based on the route headway, how early passengers show up at the stop. Two new parameters in the route choice model can be used as a part of the path utility: 1. Transfer penalty, to add inconvenience cost to each transfer, in addition to the transfer wait/walk time. 2. Fare, to incorporate the cost of boarding a transit vehicle. The value of travel time is generic at this time, but it can be individualized as needed. A skim-generation module was added, and the passenger waiting function was updated. Additional Southwest Corri- dor skim-generation code was updated in terms of Metro TDM–FAST-TrIPs integration. 5.2.2 General Transit Feed Specification Network Updates To aid DynusT and FAST-TrIPs integration, General Transit Feed Specification (GTFS) IDs were updated. C-TRAN (Clark

33 County, Washington) and SMART (Southern Metro Area Regional Transit) transit network and schedule times (GTFS) were added on the existing network (TriMet). The L35A team researched various methods to streamline this process with the aim of helping future model users reduce the time needed to prepare the transit route information for use in the modeling environment. Table 5.2 shows the transit network. Similarly, the team also checked the basic differences between FAST-TrIPs and the Emme model; all-to-all and Southwest Cor- ridor area transit skims were compared. Transit assignment cov- erage by the skims from FAST-TrIPs was analyzed by comparing it with the coverage of the existing Emme model, and to increase accessibility, additional access links were added to the existing data set to improve the accessibility for transit passengers. 5.2.3 Transit Scenario Preparation The coding for transit scenarios started from coding all routes, stop locations, and the schedule of every bus trip. Table 5.2. Transit Network Operatorsa TriMet, C-TRAN, SMART Routes 124 Trips 7,670 Stops 8,525 Stop times 417,007 a Sandy Area Metro and Canby Area Transit are not included because no general transit feed specification files were provided. New light rail transit and BRT services for the Southwest Corridor were prepared, and additional route changes on the existing routes were considered. More specifically, GTFS files were prepared for new light rail transit and BRT services, the existing three transit routes (12, 76, and 94), and the new Route 93. Developing the bus transit network for DynusT simulation was a major undertaking for a multimodal traffic simulation. GTFS files were imported into DynuStudio for editing and precise mapping of transit routes link by link onto the auto- mobile network. A diagram of bus routes is shown in Fig- ure 5.5. The script for generating transitrouteschedule.dat was also developed by the team, and the interface to the DynuStudio transit module was implemented. Buses were included in the DynusT mixed-mode DTA sim- ulation, and transit times were produced as main outputs to be fed into FAST-TrIPs. Nearly 3,200 buses were included in the transit demand for the entire simulation period. Figure 5.6 and Figure 5.7 display assigned bus volumes on the highway network and at stops, respectively. 5.3 Dynust and FaSt-trIps Integration The DynusT–FAST-TrIPs integration process involves the following steps: 1. FAST-TrIPs is run with the real demand from 11 a.m. to 7 p.m. This run will produce an output file named ft_ output_loadProfile.dat that contains information about transit dwell time. Figure 5.5. Bus routes shown in each color-coded line.

34 2. The transit dwell information produced by FAST-TrIPs can be used for the DynusT run in order to account for the delay at transit stops. However, ft_output_loadProfile.dat cannot be directly used by DynusT. A Python script is needed to convert ft_output_loadProfile.dat to TransitDwellTime.dat. 3. Next, DynusT is run for the whole period (11 a.m. to 7 p.m.) to generate the automobile skims. 4. Running DynusT for the whole period will generate an output file named AltTime_Transit.dat that contains all the transit stop time information. However, as in Step 2, this file cannot be directly used by FAST-TrIPs, and a Python script is applied to convert the contents of this file to an input file for FAST-TrIPs called ft_input_ StopTimes.dat. 5. Using the stop time information, FAST-TrIPs can then be run to produce all-to-all transit skims by applying a fake all-to-all demand for auto. At this point, the FAST-TrIPs parameters setting should be changed such that instead of using the schedule, FAST-TrIPs uses vehicle trajecto- ries. Because FAST-TrIPs runs cannot be conducted simultaneously, two FAST-TrIPs runs should be conducted Figure 5.6. Bus volumes on highway network shown in red bandwidth plot. Figure 5.7. Bus volumes at stops shown by different-sized green circles.

35 separately using the corresponding fake automobile demand to produce the midday (12 to 1 p.m.) and peak afternoon (4 to 6 p.m.) sets of transit skims. 6. Once the automobile and transit travel time skims are generated, they will be incorporated into the regional TDM to produce the final trip tables. TDM forecasts desti- nations and mode choices and produces hourly trip tables for final assignment. To conclude the process, the integrated DynusT–FAST-TrIPs model should be run once again using final hourly trip tables produced by the TDM. These steps are summarized in the flowchart shown in Figure 5.8. The integration of DynusT and FAST-TrIPs was conducted under the DynuStudio platform. The complete integration method is explained in Appendix B. 5.4 Base Model evaluation Both the DynusT and FAST-TrIPs skims were incorporated into the Metro regional TDM. The skims were verified as reasonable by Metro staff. Reasonable destination and mode choice results were anticipated from both the DynusT and FAST-TrIPs skims. The histograms in Figures 5.9 through 5.12 show travel time differences between Emme static results and DynusT–FAST- TrIPs dynamic results for zone pairs within the subarea. Given the complexities associated with the FAST-TrIPs net- works, these differences are reasonable, especially because travel time reliability is incorporated into the route choice for DynusT–FAST-TrIPs but not in the static assignments. Fig- ures 5.9 and 5.10, respectively, are for AM peak and midday automobile travel. Figures 5.11 and 5.12 are for the same periods for transit travel. Run DynusT for the whole period (11 am to 7 pm) Convert Stop Time information to FT format Run the regional travel demand model Convert Dwell Time information to DT format Run FT with real demand (11 am to 7 pm) Run FT Run FT SW fake demand (12 pm–1 pm) SW fake demand (4 pm–6 pm) Auto travel time skims Transit travel time skims Final trip tables Run integrated DT/FT model once again Figure 5.8. The DynusT–FAST-TrIPs integration framework.

36 Figure 5.9. Difference in AM peak-period SOV automobile travel times for Southwest Corridor zone pairs (weighted by trips): mean = 22.5 min, standard deviation = 4.5 min.

37 Figure 5.10. Difference in midday SOV automobile travel times for Southwest Corridor zone pairs (weighted by trips): mean = 20.4 min, standard deviation = 2.4 min.

38 Figure 5.11. Difference in AM peak-period total transit travel times for Southwest Corridor zone pairs (weighted by trips): mean 5 1.4 min, standard deviation 5 9.7 min.

39 Figure 5.12. Difference in midday total transit travel times for Southwest Corridor zone pairs (weighted by trips): mean 5 2.2 min, standard deviation 5 10 min.