Understanding the Contributions of Operations, Technology, and Design to Meeting Highway Capacity Needs (2012)

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.

52 C h a p t e r 4 This chapter presents the results of applying the enhanced DTA model features described in Chapter 2 to test the effec- tiveness of the selected operational improvement strategies shown in Table 3.2 and to confirm the applicability and useful- ness of the developed model. The model was applied in two real-world test networks. The first is a small subarea network of the Fort Worth, Texas, region and the second is a larger sub- area network within the Portland, Oregon, metropolitan area. The chapter includes a description of the network, followed by an outline of the evaluation procedures used and the results from applying the operational, design, and vehicle technology strategies in each network. The first analysis conducted within the Fort Worth network focused on individual strategy appli- cations and testing. This was in part due to the fact that all the enhancements described in Chapter 2 had to be incorporated, tested, and verified in the DTA tools and therefore could be better managed in a small network. The Portland network study entailed combinations of strategies and was intended to apply the lessons learned from the small network to a more realistic operational environment. This chapter also presents a comparison between a select set of strategies and various lane addition scenarios in order to demonstrate the concept of a strategy’s equivalent physical capacity addition. As is summarized in the next sections, it is clear—and not surprising—that the effectiveness of strategies is very much dependent on a number of contextual factors including the network congestion condition, the availability of unused capacity elsewhere on the network, the pattern of origin– destination demands (concentrated or dispersed) and the spatial and temporal extent of the strategy. Therefore there is no single “silver bullet” answer as to which strategy is most effective. What is important to note is that the tools devel- oped in this research will enable planners and engineers to directly contrast the effectiveness of nonconstruction strate- gies such as ATIS or reversible lanes or HOT lanes versus tra- ditional capacity additions on a level playing field, by not only incorporating both types in the tools but also accounting for the driver learning behavior in responding to strategies over time. Thus, in contrast to the HCM procedures, for example, which assume no demand elasticity to operational or techno- logical changes, the enhanced tools provided here allow for some elasticity to be accounted for at least in terms of route choice over time. This consideration of the time element of response also enables the introduction and generation of reliability-based measures of effectiveness into the research findings as well and begins an integration process of capacity- and reliability-based research in the SHRP 2 program. The following is a summary of key findings and conclusions from the Fort Worth network application, which is described more fully later in this chapter. • The effectiveness of each strategy cannot be quantified in a simple lookup table. The effectiveness of any particular oper- ational improvement strategy was found to be heavily depen- dent on the physical, traffic, and operating context in which it is applied. The results of the strategy applications described in this chapter are informative at a general level, but actual performance characteristics cannot be predicted for another application in a different network and/or a different context without using a tool such as the enhanced DTA model. • The effectiveness of each strategy is related to the scale (link, corridor, and/or network) at which performance is being measured. The effectiveness of strategies that modify spe- cific link characteristics (e.g., narrowing the lanes or intro- ducing reversible lanes) is likely to be most pronounced at the link level and much less so at a networkwide level. The effectiveness of strategies that broadly affect all links (e.g., improved traveler information systems) is likely to be most pronounced at the networkwide level and much less so at the link level. • The enhanced DTA model can be used to estimate the “equivalent lane addition” impact of one or more opera- tional improvement strategies. This is an especially impor- tant capability for analysts and transportation investment Strategy Testing Results and Insights

53 decision makers who must make difficult decisions about when and where to add new lanes of capacity to the exist- ing transportation infrastructure. • Important insights are gained from the enhanced DTA model on the travel time reliability effects of operational improvement strategies. Particularly in congested networks, it is significant that some operational improvement strate- gies can improve travel time reliability even if they do not materially affect the average travel time. Improving travel time reliability is an important and equally effective way of giving time back to drivers as a commensurate reduction in average travel time. It is also a way of improving the overall quality of life within a community. And yet, until now no practical method has been available to account for opera- tional strategy impacts on travel time reliability, so that this important measure has often been overlooked. • The usefulness and usability of the model will be signifi- cantly enhanced when the effects of incidents like crashes and severe weather can also be taken into account. Tradi- tional operational models do not account for the effects of events and incidents such as these. Additionally, the overall effectiveness of many operational strategies, such as those evaluated under this research effort, is incomplete if only the recurring congestion effects are considered. The following is a summary of the key findings and con- clusions from the real-world Portland network application, which is more fully described later in this chapter. • It is both feasible and practical to use the methodologies described in this report to assess alternative improvement scenarios within an urban subarea. The time and resource requirements associated with such an effort are well within the capabilities of most transportation agencies and metro- politan planning organizations. • Travel time reliability is an important performance measure to consider at both the corridor and network levels. This is particularly true in congested networks where the primary benefit from strategically placed operational improvements is improved reliability, even when average travel times are not significantly affected. • Multiple performance measures should be monitored when alternative improvement strategies are tested. Collectively, they should provide insights into capacity utilization, pro- ductivity, travel time, queuing, and reliability, which allow the user to obtain a significantly better understanding of overall impacts than would be the case if only one or two performance measures were used. • Performance measures should also be monitored at multiple spatial scales when alternative improvement strategies are tested. Specifically, performance should be evaluated at the link, corridor, and network levels, and for critical O-D pairs, to gain a complete understanding of strategy impacts and any trade-offs that might take place. Individual Strategy testing: Fort Worth Network Network Description The study network used for strategy testing is located in Fort Worth, Texas. Figure 4.1 orients the study network location relative to the Fort Worth region located within the Dallas– Fort Worth–Arlington metropolitan area. The small map on the right side of Figure 4.1 shows the Fort Worth network as coded in the DTA tool DYNASMART-P. I-35W is located in the middle of the network and provides freeway access to down- town Fort Worth, just to the north. I-35W is connected to the adjacent arterial streets by parallel frontage roads, which serve as entry and exit points to the freeway facility. As shown in the map, DYNASMART-P network has a total of 13 traffic analy- sis zones. The land use of the study area is mainly residential area except Zones 1 and 2. Land use of Zones 1 and 2 is mostly industrial or institutional. Zones 3, 6, 9, 11, 12, and 13 include some industrial activity. Therefore, the O-D pair from 1 to 2, called the Critical O-D pair (or Primary O-D pair), has the largest numbers of trips. Roadway Attributes The data set for this network was coded originally by the DYNASMART-P developers in the 1990s. The research team treated the network as an experimental framework and modi- fied demand levels to generate a congested network that would be sensitive to various strategies of interest as well as roadway features to introduce features not originally included in the net- work, such as a double left turn pockets or six-lane arterial cor- ridors. The following list provides a summary of modifications, with the number in parentheses indicating the total number of occurrences of each modification within the network. • Modified demand profile (15-minute peaking); • Modified overall LOV, HOV, and truck demand rates; • Four-way stop to actuated signal control changes (2); • Four-way stop to two-way stop changes (5); • Single lane left turn pocket to dual left turn pocket (4); and • Arterial link lane additions (8). The principal road network attributes in this test net- work are • Network data 44 Number of nodes: 180. 44 Number of links: 445. 44 Number of O-D demand zones: 13.

54 • Node control type (number of intersections including on- and off-ramps) 44 No control (on- and off-ramps): 87. 44 Four-way stop: 24. 44 Two-way stop: 6. 44 Signalized (actuated control): 59. • Traffic control data for signalized intersections 44 Two-phase control intersection: 10. 44 Three-phase control intersection: 18. 44 Four-phase control intersection: 35. 44 Max green: main 55 seconds, minor 25 seconds (or 20 seconds). 44 Min green: 10 seconds. 44 Amber: 5 seconds. All links in the study network were defined as freeway or arte- rial links, characterized by a two-regime or single-regime modi- fied Greenshields speed-flow model, respectively. Those models are explained in the DYNASMART-P user’s manual (1). The settings for each of the two facility types are as follows: • Freeway links 44 Maximum service flow rate: 2,200 pc/h/lane. 44 Saturation flow rate: 1,800 veh/h/lane. 44 Free-flow speed: 65 mph. • Arterial links 44 Maximum service flow rate: 1,800 veh/h/lane. 44 Saturation flow rate: 1,800 veh/h/lane. 44 Speed limit: 40 mph. Travel Demand Attributes Time-Dependent Demand Profile The actual analysis period is defined from 4:30 p.m. to 6:30 p.m., but in order to measure network statistics accurately Map source: © 2010 Google Maps. Figure 4.1. Fort Worth study area and DYNASMART-P simulation network.

55 over the full analysis period, vehicles were generated 30 min- utes before the analysis time period (4:00 to 4:30 p.m.) as well as 30 minutes after the completion of the analysis period (6:30 to 7:00 p.m.). Over the entire simulation period, demand levels vary every 15 minutes, as shown below in Fig- ure 4.2, with the height of each bar representing the ratio of 15-minute demand to the overall average demand. The first 30-minute period (4:00 to 4:30 p.m.) serves to load the net- work with vehicles before the start of the analysis period. The second period, or the peak analysis period (4:30 to 6:30 p.m.), has the highest demand levels and is the primary time period of interest. The third period (6:30 to 7:00 p.m.) is intended to simulate postpeak traffic and is referred to as the shoulder period. The fourth period (7:00 p.m. to network clear time) has zero demand and is included to allow sufficient time to collect statistics for all vehicles generated during the analysis period. Demand Matrices by Vehicle Class Demand for each of three vehicle types [i.e., low-occupancy vehicles (LOV), high-occupancy vehicles (HOV), and trucks] is expressed in twelve 15-minute O-D trip tables for the 3-hour demand period. The vehicle trip combinations used in the sim- ulation network are LOV (85.0%), HOV (9.5%), and trucks (5.5%). LOV and HOV were assumed to have occupancy rates of 1 and 2.3 passengers per vehicle, respectively, while trucks use passenger car equivalent factors that vary based on the input roadway grade. Demand Pattern for Alternate Baseline Network Although demand levels were calibrated to generate a reason- ably congested network, this network included only a small percentage of trucks. To test some truck-related strategies a modified high level of truck demand was simulated. Thus, rather than bias the results of the majority of the non-truck strategies tested, an alternate baseline was generated. The modi- fied demand produces approximately 2,500 trucks on the major southbound O-D (from 1 to 2) and 550 trucks on the major arterial southbound O-D (from 3 to 12). User Classes User classes are defined in terms of access to travel information. The default user class, termed the Unequipped Class, has no access to real-time information and must base all route choice decisions on the learning methodology described in Chapter 2. However, two additional user classes, the Pretrip Information Class and the En Route Information Class, were developed to establish sensitivity to real-time information-based strategies. Each of these classes has access to a snapshot of current travel times along potential routes, either before departure, as with the Pretrip Information Class, or continuously along the route, as with the En Route Information Class. Within the baseline, 98% of drivers have no access to information (Unequipped), 1% can access pretrip information only (PT), and 1% can access con- tinuous en route information (ER). 0.71 0.76 0.80 1.09 1.21 1.11 1.06 0.96 0.87 0.90 0.82 0.77 0.00 0.00 0.00 0.10 0.20 0.30 0.40 0.50 0.60 0.70 0.80 0.90 1.00 1.10 1.20 1.30 4:00~4:15 4:15~4:30 4:30~4:45 4:45~5:00 5:00~5:15 5:15~5:30 5:30~5:45 5:45~6:00 6:00~6:15 6:15~6:30 6:30~6:45 6:45~7:00 7:00~7:15 7:15~7:30 Time Period 1 Period 2 Period 3 Period 4 R a tio Figure 4.2. Baseline network demand profile.

56 The day-to-day learning procedure described in Chapter 2 was implemented within DYNASMART-P to model the ways in which drivers choose routes on a daily basis, as well as how they learn from previous travel experiences. The method establishes a minimum travel time improvement threshold, and drivers compare their default paths (originally based on minimum travel time) against travel times on alternate paths during preceding days. In order to simulate the gradual way in which drivers change their daily routine, only 15% of drivers in the Unequipped Class are permitted to have the option to switch their preferred path each day based on obser- vations over the previous 5 days. All information users (PT and ER), on the other hand, were given the option to update their preferred paths on a daily basis. The “switching rate” refers to the actual total daily percentage of drivers who decide to take an alternate path based on this information, thus mod- ifying their preferred path, and the variation of this value between 0% (minimum) and 17% (15% + 1% + 1% maxi- mum) provides an indication of day-to-day network stability. Table 4.1 summarizes these thresholds. Strategies Evaluated A comprehensive review was undertaken of technologies affecting traffic operations. An inventory was also developed of operational, design, and technological strategies and tactics for achieving improvements in sustained service rates for freeways and/or arterial segments. Nearly 100 strategies and tactics were identified. A subsequent ranking process accounted for the effect of each strategy/tactic on road segment capacity; whether the particular strategy/tactic is in the purview of decision makers to implement; and barriers to actual implementation. On this basis, the approximately 100 initially identified strate- gies and tactics were distilled to the 25 presented in Table 3.2 and considered to be most promising for actual application. Measure of Effectiveness Each of the strategies tested represents an attempt to mitigate congestion and improve the productivity of the network, but not all could be expected to generate results significant at the network level. It was therefore necessary to compare results at a relative scale of interest. For example, with approximately 25% of the networkwide demand occurring on the south- bound freeway facility, it was reasonable to examine network- wide results for many of the freeway-based strategies. However, many of the effects of the arterial strategies were only signifi- cant at the corridor level, requiring analysis at this scale. Some strategies dealt with individual intersections, making link- based analysis most appropriate. Therefore, although the base- line provided a standard set of comparison results, the scale of analysis was determined on a case-by-case basis. A variety of performance measures are available at the link, corridor, O-D, and networkwide levels from DYNASMART-P output. Addi- tionally, freeway bottleneck summary information is provided as a potential diagnostic tool for practitioners. Link-Level Performance Measures For each link (i) in DYNASMART-P, the following MOEs are reported for every 15-minute interval over all simulation days: • Link vehicle count (veh/15 minutes); • Average travel time (minutes); • Space mean speed (mph); • Vehicle density (veh/mi/lane); • Queue length (defined as the ratio of vehicle queue length to the link length); • Freeway link breakdown indicator (for a freeway link, if at the end of every 15-minute interval, there is a queue, 1; otherwise, 0); • Number of cycles with a queue at the start of the red phase (exclusively for signalized links); and • Link capacity (veh/h/lane). Corridor-Level Performance Measures Several corridor MOEs such as density (veh/mi/lane), space mean speed (mph), queue length on corridor, travel time, and breakdown count or cycle failure count for a user-specified Table 4.1. User Class Settings User Class Unequipped PT ER Day-to-day learning Percentage of total vehicles 98% 1% 1% Daily learning rate (maximum switching %) 15% 100% 100% Daily learning improvement threshold (minutes) 5 minutes 5 minutes 5 minutes Maximum daily learning percentage 17% Daily path selection Pretrip path improvement threshold (minutes) — 2 minutes — En route path improvement threshold (minutes) — — 2 minutes

57 corridor can be created by aggregating one or more link- based MOEs. Network-Level Performance Measures At the network level, DSP generates average travel time (minutes/veh) for each of the vehicle/user subgroups listed below: • Networkwide (all vehicles); • Low-occupancy vehicle (LOV) group; • High-occupancy vehicle (HOV) group; • Unequipped vehicle group (vehicles that have no access to pretrip or en route information); • Pretrip information vehicle group; • En Route information vehicle group; and • Critical O-D vehicle group. The average travel time for each subgroup on a single sim- ulation day is also reported for the entire simulation period. Users may also evaluate this MOE over multiple simulation days based on their needs. Simulation Procedure In order to appropriately compare a baseline (unmodified net- work which represents current condition) to a strategy applica- tion case (modified network with enhancements), the user should follow these steps: • Simulate the baseline network for 200 days by using the baseline O-D Demand Matrices. • Export vehicle and path information. • Simulate the baseline network for an additional 50 days by using the vehicle and path files exported from the original run. • Simulate the modified network for an additional 50 days by using the vehicle and path files exported from the origi- nal run. • Compare the baseline and the strategy results for the final 20 days of the simulation period. • Subsequent applications suggest that this analysis time horizon is more than adequate even for larger networks, but the number of simulation days is a matter that can be individually judged at the time of each application. Definition of Simulation Analysis Regimes Figure 4.3 illustrates the defined simulation regimes that are defined to compare the results appropriately. The figure shows daily networkwide average travel time and daily route switch- ing rate from the baseline and the strategy results. Regime I: Baseline Stabilization Period Before strategy testing can begin, the model must first allow drivers to learn the network and establish their preferred paths, just as anyone starting a new job may try several dif- ferent routes initially in search of their preferred route to work. The daily switching rate, as noted, provides an indica- tion of the overall stability of the network. The network was therefore allowed to run until relatively stable conditions were established, characterized by no observable general trend in the daily switching rate. Using 200 days provides a conservative estimate of this value, as shown for Regime I in Figure 4.3. Figure 4.3. A 250-day simulation process.

58 Regime II: Strategy Stabilization Period To test the effects of any type of network modification, the user must (a) allow drivers a sufficient number of learning days to adjust to the modified network and learn new paths as neces- sary before collecting statistics and (b) compare these results against a set of baseline runs generated by using the same sequence of random numbers. To do so, the user must first load the initial 200-day baseline results and change the settings from O-D-based vehicle generation to path-based vehicle genera- tion to simulate the network by using the learned paths from the baseline stabilization period. Regime III: Results Comparison Period By simulating a modified network in parallel to the unmodi- fied baseline network for 50 days, as discussed above, the final 20 days can serve as the stabilized comparison period that follows the same random number sequence, thus allowing for a correct protocol for comparison. As such, all strategy tests should follow the same procedure outlined above, comparing the final 20 days of simulation (Days 231 to 250) against the corresponding baseline simulation days. Results and Key Insights Baseline Results The simulation results from the baseline, especially link vol- umes and speeds, can be used for the initial network calibra- tion. Networkwide travel time can also be used for examining whether the developed network produces reasonable results. Networkwide average summary statistics for Days 231 to 250 of the baseline user class are shown in Table 4.2. Link-based simulation performance measures from the baseline such as speed, density, and queue length can be used for diagnosing the network to find a potential starting point for strategy application. In addition to those, DYNASMART-P provides a bottleneck diagnosis function and statistics on the number of cycles with a residual queue at the start of the red phase on signalized links. Both can be reported in text output and in a visualized map as well. The results from the bottle- neck diagnosis function are shown below. For each freeway, on-ramp, and off-ramp link, the following information is reported for each day: • Total number of bottleneck-delayed vehicles (veh); • Total bottleneck-caused delay (minutes); and • Average bottleneck delay per vehicle (minutes). Table 4.3 provides a 20-day average (Days 231 to 250) of the five bottlenecks that generate the largest amount of delay in the baseline network, with the location of each depicted in Figure 4.4. To improve these bottleneck sections, some strate- gies were implemented and evaluated. Assessing the Effectiveness of a Freeway Reversible-Lane Strategy In the course of the research, many of the 25 strategies pre- sented in Table 3.2 were implemented in the Fort Worth net- work to assess their effectiveness in maintaining SSRs for freeways and/or arterial segments. In this section, the freeway Table 4.2. Baseline Networkwide Summary Statistics (Days 231–250) Class Overall Unequipped Pretrip En Route LOV HOV Critical O-D (1➔2) Average travel time (minutes) 7.49 7.50 6.94 7.00 7.49 7.53 9.58 Table 4.3. 20-Day Average of Baseline Bottleneck Information (Days 231 to 250) Bottleneck ID Link Number Type Number of Vehicles Delayed Total Delay (minutes) Average Delay (minutes) Frequency of Activation (% of 20 Days) 1 69 Freeway 3,907 14,063 1.90 100 2 89 Freeway 3,858 14,053 1.86 85 3 94 Freeway 4,457 10,605 1.77 100 4 86 Freeway 2,348 8,568 1.56 75 5 117 Freeway 1,618 5,585 1.04 65

59 reversible-lane strategy was selected to demonstrate how to implement strategies in a DYNASMART-P network and what kind of results can be extracted from the simulation outputs. In order to measure the effects of the strategy, results from the baseline and strategy are compared. Reversible lanes, or counterflow lanes, are lanes that allow traffic to flow in either direction through the use of dynamic message signs or movable barriers. A reversible lane is typi- cally used to improve traffic flow during peak periods with highly directional flow. The team implemented this strategy to improve the apparent freeway bottleneck in Sections 1, 3, and 4, which are illustrated in Figure 4.4. The I-35 West facility in the Fort Worth network has four lanes over the entire 3.1-mi corridor, as highlighted in Fig- ure 4.5. It represents the peak direction of travel during the evening peak period. The corridor shown includes three on- ramps and five off-ramps. First, a lane was added to each of the nine links in the southbound direction so that reduction factors within DYNASMART-P could be used to simulate the opening and closing of lanes. Nine southbound links were modified to simulate a reversible-lane scenario with one northbound lane converted to a southbound lane during the peak hour. This equates to a 25% capacity reduction during the peak hour in the northbound direction, since four travel lanes are reduced to three by this strategy. The lane reduction in the off-peak northbound direction begins at Minute 45, continuing through Minute 135. The corresponding lane addition (lane opening) occurs between Minutes 60 and 120, where the reduction factor is removed to allow five full lanes of travel. Two 15-minute periods before and after the reversible-lane operation was to take effect were considered as the clearance time periods. Network-Level Results Comparison At the network level, although there was an average reduction in total breakdown count, bottleneck vehicle count, and vehi- cle delay caused by bottlenecks, overall average travel time actually increased by about 6%. Although networkwide travel time is highly variable due to the use of stochastic capacity, this travel time increase may suggest that the benefit to the southbound direction of an additional travel lane was out- weighed by the substantial disbenefit caused to traffic in the northbound direction. The bulleted list that follows summa- rizes these results, with the number of standard deviations of improvement shown to the right. The vertical bar () rep- resents the performance of the baseline. The star (*) repre- sents the standard deviation of improvement. The number of stars represents the size of the improvement in terms of the number of standard deviations of the performance measure. If stars show up on the right side of the bar, it means the per- formance measure associated with the strategy shows an improvement. If they show up on the left side of the bar, it means the performance measure has degraded. The average and 95th percentile values for several MOEs of interest are depicted in Table 4.4. Figure 4.4. Visualized bottleneck locations. Figure 4.5. Reversible-lane implementation in Fort Worth network.

60 Network Performance: • Travel time * • Breakdown count ** • Bottleneck vehicle count  • Bottleneck delay  Corridor-Level Results Comparison At the corridor level, the additional southbound capacity gener- ated an increase in processing efficiency during the peak 15 minutes in the southbound direction and attracted a greater number of vehicles from adjacent routes to the freeway corridor during the peak period. All primary performance measures showed an improvement from this strategy, including reduc- tions in travel time, density, queue length, breakdowns, and bottleneck delay. Despite these gains, service degradation in the northbound direction was both significant and severe. Speed decreased sig- nificantly, corresponding to a significant increase in travel time and density. In short, the northbound direction did not have enough spare capacity to sacrifice a full lane to the south- bound direction during the peak hour, leading to a significant decrease in quality of service with the reversible-lane in opera- tion. The bulleted list below summarizes these results. Average and 95th percentile values are presented for each MOE of interest in Table 4.5. Table 4.6, Table 4.7, Table 4.8, Figure 4.6, and Figure 4.7 provide illustrations of speed by link along the corri- dor. It should be noted that bottleneck vehicle counts and bottleneck delay are summarized for the entire simulation period, while other MOEs are presented only for the peak 15 minutes. Southbound Direction: • Speed * • VMT * • Travel time * • Density * • Queue length * • Breakdown count * • Bottleneck vehicle count * • Bottleneck delay * Northbound Direction: • Speed *** • VMT ** • Travel time * • Density  • Queue length * • Breakdown count  • Bottleneck vehicle count * • Bottleneck delay * Table 4.4. Comparing Reversible-Lane Performance at the Network Level MOE Baseline Reversible-Lane Strategy Average 95th Percentile Average 95th Percentile Change in Average (%) Travel time (minutes) 7.72 9.16 8.22 9.23 6.4 Breakdown count 24 29 18 23 -25.2 Bottleneck vehicle count 26,865 35,784 25,704 33,124 -4.3 Bottleneck delay (hours) 1,314 2,456 1,293 2,111 -1.6 Note: MOE = measure of effectiveness. Table 4.5. Reversible-Lane Peak 15-Minute Results (Freeway Corridor: Southbound) MOE Baseline Reversible-Lane Strategy Average 95th Percentile Average 95th Percentile Change in Average (%) Speed (mph) 56.94 62.53 63.56 64.12 12 VMT (veh * mi/15 minutes) 5,153 5,787 5,809 6,109 13 Travel time (minutes) 5.17 10.61 3.70 3.72 -28 Density (pc/mi/lane) 31.48 59.08 19.62 20.80 -38 Queue length (mi) 0.15 0.36 0.04 0.09 -73 Breakdown count 0.95 2.00 0.15 1.00 -84

61 Table 4.6. Reversible-Lane Peak Daily Bottleneck Information (Freeway Corridor: Southbound) MOE Baseline Reversible-Lane Strategy Average 95th Percentile Average 95th Percentile Change in Average (%) Bottleneck vehicle count 7,266 12,858 1,073 4,981 -85.2 Bottleneck delay (hours) 391 1,541 32 134 -91.7 Table 4.7. Reversible Lanes Peak 15-Minute Results (Freeway Corridor: Northbound) MOE Baseline Reversible-Lane Strategy Average 95th Percentile Average 95th Percentile Change in Average (%) Speed (mph) 56.19 61.30 47.10 53.81 -16 VMT (veh*mi/15 minutes) 5,454 5,669 4,940 5,271 -9 Travel time (minutes) 6.08 10.60 7.98 12.86 31 Density (pc/mi/lane) 39.53 69.61 44.89 70.05 14 Queue length (mi) 0.33 0.57 0.17 0.32 -48 Breakdown count 1.10 3.00 1.15 2.00 5 Table 4.8. Reversible-Lane Peak 15-Minute Results (Freeway Corridor: Northbound) MOE Baseline Reversible-Lane Strategy Average 95th Percentile Average 95th Percentile Change in Average (%) Bottleneck vehicle count 7,483 13,171 14,515 24,724 94.0 Bottleneck delay (hours) 440 1,251 905 1,661 105.6 Speed vs. Distance 0.00 10.00 20.00 30.00 40.00 50.00 60.00 70.00 0 5000 10000 15000 20000 25000 Sp ee d (m ph ) Distance (ft.) Baseline Reversible Lanes Baseline Average Strategy Average Figure 4.6. Southbound freeway corridor peak: 15-minute speed versus distance.

62 Thus, for this particular network, directional peaking along the freeway was not significant enough to warrant the removal of a lane from the off-peak direction to provide additional capacity to the peak direction. It should be noted, however, that the strategy did provided significant benefits to the peak direction, and for cases with lighter flow in the off-peak direction, or where the peak direction is highly critical for network performance, the strategy could have proved to be successful. Thus, the context in which the strat- egy is applied can have a significant effect on its overall effectiveness. Strategies Involving Equivalent Physical Capacity Addition Concept In addition to comparing measures of effectiveness across multiple strategies, many practitioners would find it espe- cially useful to be able to compare operational strategies to a typical lane addition scenario. In this manner, they would be able to determine the “equivalent capacity gain” of the candidate strategy. Figure 4.8 illustrates one possible way to do this through a hypothetical relationship between lane miles added to a network and performance, measured in terms of total network travel time. It may be useful to develop such a relationship from a variety of travel demand, HCM, or simulation models for a given network case study as lane mile additions represent the traditional way of increasing network capacity. It is important to note, how- ever, the relationship shown in Figure 4.8 does not take into account the effect of latent demand on performance, nor is it necessarily continuous. Both of these issues are worthy of further exploration. Even so, the ability to implement the concept presented in Figure 4.8 would represent a signifi- cant step forward for the transportation profession and would greatly improve the quality and impact of informa- tion available to transportation investment decision makers. A set of nonconstruction improvements that reduces net- work travel times from A to B in Figure 4.8 is effectively equivalent to adding D minus C lane miles, or a construction- based capacity increase of (D - C)/C%. A method for under- taking the implementation of this concept is described in the following paragraphs. Implementation In order to illustrate the equivalent capacity gain concept for a few selected strategies, three physical capacity addi- tion scenarios were applied to the primary freeway corridor Speed vs. Distance 0.00 10.00 20.00 30.00 40.00 50.00 60.00 70.00 0 5000 10000 15000 20000 25000 Sp ee d (m ph ) Distance (ft.) Baseline Reversible Lanes Baseline Average Strategy Average Figure 4.7. Northbound freeway corridor peak: 15-minute speed versus distance. Effect of Non-Construction Improvements on Network Travel Time N et w or k Tr av el T im e Equivalent Capacity Gain Added Network Lane Miles A B C D Figure 4.8. Equivalent capacity gain concept.

63 and tested in DYNASMART-P. As depicted in the map in Fig- ure 4.9, lanes were added on I-35 Southbound, the primary path for O-D pair 1-2. The four illustrations in Figure 4.9 show the baseline condition and the locations of the lane additions for each of the three lane addition scenarios, A, B, and C. Scenario A involved adding one lane to the exist- ing four lane segments, creating a continuous five-lane freeway corridor with no lane additions or reductions. In Scenario B, one complete lane was added throughout the entire corridor, maintaining the 0.9-mi segment with an additional lane. In Scenario C, the freeway was converted to a six-lane facility over the entire length of the corridor. The total lane miles added for Scenarios A, B, and C were 7.6, 8.5, and 16 lane miles, respectively. This is equivalent to 22%, 24%, and 46% lane mile increases for Scenarios A, B, and C, respectively. Four operational improvement strategies were selected for comparison, involving three separate enhancement catego- ries: pretrip information (technological), en route informa- tion (technological), additional narrow lanes in the existing cross section (design), and reversible lanes (operational) for improving identified bottleneck Sections, 1, 3, and 4 in Figure 4.4. The following sections present the comparison results associated with this particular application of the iden- tified improvement strategies and are of course subject to the caveat that the learning model and strategy effectiveness models are representative of real-world behavior. Thus, while the general trends will hold, specific MOEs may be subject to some variation. Day 1 Results Examination of Day 1 results represents a traditional evalua- tion method for strategy testing with no learning or changes in demand. In other words, the examination of Day 1 results gives the analyst the opportunity to view the consequences of implementing the improvement strategies before drivers have a chance to respond either temporally or spatially, similar to what a highway capacity analysis procedure would yield. For obvious reasons, O-D pair 1-2 was selected for the travel time analysis. As shown in Figure 4.10, the ATIS pretrip information and ATIS en route information strategies gener- ated the same results as the baseline scenario, as both strate- gies rely on changes in route selection over time to have any meaningful effect. The remaining strategies exhibited signifi- cant improvement over the baseline scenario, with approxi- mately a 5-minute travel time reduction for each. As capacity increased along the corridor for each strategy with no increase in demand, all vehicles were able to travel at free-flow speed, generating approximately the same results for each strategy. Days 31 to 50 Results A significant contribution of this research to the analysis of operational strategies was the addition of a driver learning algo- rithm which was implemented within the DYNASMART-P environment. Drivers are therefore able to respond to changes in the network and gradually learn the most efficient routes for a given O-D and departure time. To allow drivers time to respond to changes in the network, each scenario was A Baseline # of Lanes B C 4 5 5 5 5 6 6 4 +16.0 (+46%)+8.5 (+24%) Lane-Mile Addition +7.6 (+22%) 8.5 miles 1.9 miles1.9 miles 0.9 miles0.9 miles 5.7 miles5.7 miles 8.5 miles Figure 4.9. Illustrations of three physical lane addition scenarios.

64 simulated for a total of 50 days. The results depicted in the following exhibits represent average values from Days 31 to 50 to take into account supply stochasticity. Standard error bars are shown where appropriate. The following MOEs were selected for comparison: • Networkwide 44 Mean travel time. • Primary O-D (1–2) 44 Mean travel time; 44 95th percentile travel time; 44 Travel time index = mean travel time/free-flow travel time; and 44 Buffer index = (95th percentile travel time - mean travel time)/mean travel time. Figure 4.11 shows the networkwide mean travel times for all scenarios. Although both ATIS strategies reduced network- wide travel times, the narrow lanes strategy was the most effective among the nonconstruction alternatives. In all three cases however, each of the lane mile addition scenarios proved to be a more effective alternative. The reversible-lane strategy, on the other hand, increased networkwide travel times largely due to added congestion in the northbound direction caused by the peak-hour lane reversal. It is important to note that in addition to travel time savings, each of the three effective strat- egies also decreased the travel time variability on the primary O-D. This concept of reliability is of critical importance to the commuter, and an increase in reliability is a benefit largely ignored when only measuring average travel times. For each of the lane mile addition scenarios, reliability was increased to the point that variability in travel times all but disappeared entirely. Figure 4.12 shows travel times along the primary O-D (1-2) for each of the evaluated scenarios. All strategies reduced O-D travel times relative to the baseline, including the reversible- lane strategy, as the primary O-D includes only travel times in the peak (southbound) direction. Overall, the narrow-lane strategy was the most effective in reducing travel times in the primary O-D, although the strategy was slightly less effective Tr av el T im e (M inu tes ) 16 14 12 10 8 6 4 2 0 B as el in e A TI S Pr e- tri p A TI S En -ro ut e N ar ro w L an es R ev er si bl e La ne s A (+ 22 % ) B (+ 24 % ) C (+ 46 % ) Figure 4.10. Mean travel time-Day 1-primary O-D 1-2. Figure 4.11. Average of Days 31 to 50 networkwide travel time. Ba se lin e AT IS P re -t rip AT IS E n- ro ut e N ar ro w L an es Re ve rs ib le L an e A (+ 22 % ) B (+ 24 % ) C (+ 46 % ) 0 2 4 6 8 10 12 14 16 Tr av el T im e (M inu tes ) Mean + Standard error Mean - Standard error * Note: asterisk = accounts for added congestion effect on northbound freeway corridor due to lane reversal.

65 than lane mile addition Scenario A. Error bars indicate that each of the strategies decreased the travel time variability for the primary O-D, indicating an overall increase in reliability. The 95th percentile travel times for the primary O-D, as shown in Figure 4.13, also demonstrate this trend. Figure 4.14 provides the travel time index for each of the evaluated scenarios on the primary O-D. The baseline sce- nario has a travel time index of around 1.7, which is to say that average travel times for the primary O-D were approxi- mately 70% higher than under free-flow conditions. The travel time index decreased for each of the scenarios, and the narrow-lane strategy was able to decrease travel times to less than 30% greater than free-flow travel times during the peak hour. Although each of the lane mile addition scenarios gen- erated lower travel time index values than the narrow-lane strategy, even the most expensive option (Scenario C) was only able to reduce the travel time index to slightly less than 1.2, highlighting how the method could be used in a cost- benefit analysis of potential strategies. The buffer index shown in Figure 4.15 provides one method of evaluating travel time reliability for a given O-D or corridor of interest. Small buffer index values indicate low variability of travel times, or high reliability. As dis- cussed previously, the buffer index clearly demonstrates that each of the scenarios led to an overall increase in reli- ability. Interestingly, however, the buffer index reveals that the ATIS pretrip strategy was the most effective of the four nonconstruction alternatives at increasing travel time reli- ability. Such an analysis allows the practitioner to evaluate strategies based on a number of potential objectives and perform a network-appropriate customized approach to congestion mitigation. Note: asterisk = effects in peak direction with lane addition only. Ba se lin e AT IS P re -t rip AT IS E n- ro ut e N ar ro w L an es Re ve rs ib le L an e A (+ 22 % ) B (+ 24 % ) C (+ 46 % ) 0 2 4 6 8 10 12 14 16 Tr av el T im e (M in ut es ) Mean + Standard error Mean - Standard error * Figure 4.12. Travel times for trips serving primary O-D 1-2. Note: asterisk = effects in peak direction with lane addition only. Ba se lin e AT IS P re -t rip AT IS E n- ro ut e N ar ro w L an es Re ve rs ib le L an e A (+ 22 % ) B (+ 24 % ) C (+ 46 % ) 0 2 4 6 8 10 12 14 16 Tr av el T im e (M in ut es ) * Figure 4.13. 95th percentile travel times for trips serving primary O-D 1-2.

66 Limitations and Cautions The enhanced DTA model developed through this effort and described in the preceding chapters for a test network provides a practical methodology for assessing the ability of various operational strategies, either singly or in combina- tion with one another, to forestall or eliminate the need to construct additional lane miles of capacity within a trans- portation network. The methodology provides effectiveness assessments about both travel time and reliability at the link, corridor, and network levels. It is already integrated into two dynamic traffic assignment (DTA) modeling pro- cedures (including the open source program DTALite as well as DYNASMART-P) and can be integrated into others as well. Unfortunately, the usefulness of the methodology is con- strained by the fact that the modeling environments in which it operates do not account for crashes, weather, and other events that can cause nonrecurring congestion. Nonrecurring congestion represents a significant part of the delay and frus- tration experienced by travelers and therefore has a substan- tial impact on both travel time and reliability. Yet, traditional operational models do not typically account for its effects. Additionally, the overall effectiveness of many operational strategies, such as those evaluated under this research effort, is misrepresented if only the recurring congestion effects are considered. As an example, the use of narrow lanes as an oper- ational improvement strategy will typically result in more capacity, reduced travel time, and improved reliability in an environment where incident effects are not considered. But narrow lanes might also increase the potential for crashes, thereby reducing their effectiveness if nonrecurring conges- tion effects are taken into account. As another example, ramp metering does not add much in terms of capacity, but it Note: asterisk = effects in peak direction with lane addition only. Ba se lin e AT IS P re -t rip AT IS E n- ro ut e N ar ro w L an es Re ve rs ib le L an e 0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0 Tr av el ti m e in de x A (+ 22 % ) B (+ 24 % ) C (+ 46 % ) * Figure 4.14. Travel time index for trip serving primary O-D 1-2. Note: asterisk = effects in peak direction with lane addition only. Ba se lin e AT IS P re -t rip AT IS E n- ro ut e N ar ro w L an es Re ve rs ib le L an e A (+ 22 % ) B (+ 24 % ) C (+ 46 % ) 0.0 0.1 0.2 0.3 0.4 0.5 0.6 B uff er in de x * Figure 4.15. Buffer index for trips serving primary O-D 1-2.

67 stabilizes flow, lowers the probability of breakdown during demand surges, and reduces the potential for crashes. In this case, the benefit of ramp metering will be underestimated unless nonrecurring congestion effects are also considered. Therefore, a comprehensive assessment of the overall effective- ness of these operational strategies needs to account for their effects on both recurring and nonrecurring congestion. Much value will be gained from future efforts that extend the capability of the DTA methodology to include an assess- ment of nonrecurring congestion effects, incorporating this additional capability into at least one functional modeling environment. Illustrative Application of Methods, Metrics, and Strategies The overall net impact of a combination of operational improvement strategies deployed within a subarea or network will almost always be different from the sum of their individual effects, due to the influences each implemented strategy has on the others. The ability of a mesoscopic model to take these interactions into explicit account is one of the important advantages that come from implementing the methodological enhancements of this project in such an environment. Combinations of various operational improvements strat- egies identified in this project were applied to a subarea of the Portland network in order to demonstrate the feasibility of this approach. In the process, useful insights were obtained that may have application in a broader context. The subarea network selected for these purposes is identi- fied in Figure 4.16. It was selected for this prototype applica- tion for a number of reasons: • It is relatively large in size and therefore represents a good opportunity to test scaling issues associated with the method applications. • It includes both north-south and east-west freeway seg- ments, as well as multiple interchanges on both segments. • It includes a full range of arterial streets that are generally organized around a grid pattern. Consequently, there are good opportunities within this particular subarea to conduct Figure 4.16. Portland network study area.

68 separate analyses of individual links, of separate corridors, and of the subarea network as a whole. The area is characterized by significant congestion on both freeway and arterial segments during typical weekday eve- ning peak hours. These conditions present opportunities to test solution alternatives ranging from new construction to a variety of operational improvement strategies. The purpose of this application was to demonstrate, through the detail of an actual and real-world example, the manner in which each step in the procedure can be undertaken; the thought process that accompanies each step and informs deci- sions regarding next steps; and a representative range of findings/conclusions that can be anticipated as outcomes of the procedure. The application was developed specifically for the purposes of this report and was not used to inform actual investment decisions in the Portland metropolitan area. There are references throughout this chapter to DTALite and DYNASMART-P, both of which are mesoscopic dynamic traffic assignment models. DTALite is a fully functional, open source dynamic traffic assignment model that can be down- loaded without charge from http://sourceforge.net/projects/ dtalite/. DTALite incorporates all the DTA model enhance- ments developed through this project. DYNASMART-P, Version 1.3.0, is available for a charge from the McTrans Center at http://mctrans.ce.ufl.edu/featured/dynasmart. Ver- sion 1.3.0 does not include any of the model enhancements developed through this project effort. The two models were used together in the execution of this prototype application, but it should be emphasized that neither model is essential to the implementation and execution of the enhanced analytic and diagnostic methods described herein. These particular models were chosen for use in this particular application because of the research team’s familiarity with them, and also to maintain consistency with the platforms used in the development and testing of the model enhance- ments and diagnostic tools described earlier. Other DTA mod- els are available and could also be used in lieu of either DTALite or DYNASMART-P as the user prefers, providing the model enhancements and diagnostic tools developed within this project are appropriately integrated into them. Description of the Subarea The subarea that was selected for this prototype application, and illustrated in Figure 4.16, is located in the southwestern part of the Portland metropolitan area. It encompasses a fairly large area and includes facilities that have statewide, regional, and/or local significance. • Highway 26, also known as the Sunset Highway, forms the northern boundary of the study area. It is the primary connector to an area known as the Silicon Forest for its high technology industry employment and surrounding residen- tial population. Beyond the subarea’s western boundary, Highway 26 continues westward for about 70 miles to the Oregon Coast and therefore is used extensively for freight movement as well as tourism and recreation. It is classified as a road of statewide importance because of the scope and scale of activities that rely on it. • The eastern boundary of the subarea is defined by a portion of Highway 217, which connects between Interstate 5 (I-5) on the south and Highway 26 on the north. Therefore, in addition to serving travel demands within the corridor in which it is located, Highway 217 also represents an important connection for travelers from the south who are destined to points in the western part of the study area and vice versa. • The southern boundary of the study area is defined by Farmington Road, which travels in a northeast-to- southwest direction and represents an effective boundary line between the western and southwestern parts of the Portland metropolitan area. Farmington Road transitions from an urban to a rural environment as it moves to the south and west, and in fact it is located outside the Urban Growth Boundary at the westernmost end of the subarea. • The western boundary of the study area is defined by River Road, SE 10th Avenue, and NE Brookwood Parkway in such a way as to stay just to the east of the majority of Hillsboro’s downtown core area and to connect between Farmington Road and Highway 26. The interior of the subarea comprises a surface street sys- tem that includes a grid-like arterial road system for both east-west and north-south travel. Tualatin Valley Highway, also referred to as TV Highway or State Route 8, is the pre- ferred route for east-west arterial travel and includes frequent bus service. It is also a fairly congested route, particularly dur- ing weekday peak hours. Table 4.9 presents summary statistics about the Portland subarea network used in this application. It is a reasonably large Table 4.9. Summary Statistics about the Portland Subarea Network Network Characteristic Entire Network Subarea Network Number of traffic analysis zones 2,013 208 Number of nodes 9,905 857 Number of links 22,748 1,999 Number of originating vehicle trips 1.2 million 212,000 Average travel time (evening peak period) 22 minutes 10.5 minutes

69 subarea with more than 200 traffic analysis zones (TAZs) and more than 200,000 originating vehicle trips during the 4-hour weekday time period (3 p.m. to 7 p.m.) that is analyzed. This subarea is well suited to the purposes of this demonstra- tion effort due to its size, the diversity and redundancy of vari- ous facility types, and the prevalence of congested conditions during typical weekday evening peak-hour conditions. Building and Calibrating the Portland Subarea Network An initial version of the Portland metropolitan area network was provided to the team in the same VISUM model format that Metro staff uses for their current travel demand forecast- ing activities. Metro staff also provided additional network detail from a variety of other sources, including signal control information that had been assembled some years earlier as part of their experimental work with the TRANSIMS model; detailed information about approach lane configuration; and the location and length of turn pockets. These data items were easily ported and transformed into a standard format that is used by multiple DTA programs (including DTALite, DYNASMART-P, and DynusT). After the initial data sets were received and transformed, it was found to be necessary to conduct a variety of error- checking activities on the network. DTA models interact with the network differently from traditional travel demand mod- els like VISUM, and so some network elements that do not cause problems in a VISUM analysis can still represent incon- sistencies that need to be rectified for a DTA analysis. Exam- ple issues that were discovered and corrected through this process will help to clarify the kind of examination and cor- rective effort that is typically needed: • Some centroid connectors from the VISUM model were found to be tied directly into real intersections. This will cause unrealistic operating conditions for the simulated intersection in a DTA model environment. To resolve this problem, the tie-in point for these centroid connectors was relocated to a mid-block location. • Some TAZs had only one or two centroid connectors, caus- ing too much volume to enter the network via a single street. More centroid connectors were added to resolve this problem, so the travel demand was dispersed more appro- priately across adjacent road segments. • The speed, length, and capacity of centroid connectors were sometimes defined within the VISUM model in ways that could significantly affect the overall travel time results for the network and/or corridors being investigated. The remedy to this problem was to adjust these parameters to be more representative of actual driving conditions on the local and collector street system inside each TAZ. • The network structure at the subarea boundaries needed careful examination to ensure realistic performance charac- teristics during the DTA model runs. An example is shown in Figure 4.17. Here, the southernmost entry node for north- bound traffic on Highway 217 initially connected to another mainline node on Highway 217, from which vehicles could either continue northbound on Highway 217 or exit onto an off-ramp (see Figure 4.17a). However, testing revealed that any backups from the off-ramp onto the mainline link would also block all northbound through traffic and cause unreal- istically long delays to the through-traffic component of the entering traffic. This problem was resolved by reconfiguring the entry node so that it connected directly to two possible destinations: (a) the off-ramp and (b) the northbound through lanes on Highway 217. This is shown schematically in Figure 4.17b. It is noteworthy that some inaccuracies (relative to true existing conditions) were purposely introduced into the net- work structure to ensure that this exercise is used only to (a) (b) Exit to TV Highway Network Entry Node NB Highway 217 Mainline Exit to TV Highway Network Entry Node NB Highway 217 Mainline Figure 4.17. Network coding: (a) initial network coding and (b) modified network coding.

70 demonstrate the new methods and model capabilities. Thus, for example, the number of through lanes on some arterials was modified, as was the length of some left and right turn pockets. These inaccuracies do not affect the overall value of this demonstration exercise but do help assure that the results are not used to inform actual public investment decision making or policies. The entire Portland area network was initially simulated by using DYNASMART-P and DynusT. This was done for a 4-hour time period (3 p.m. to 7 p.m.) across multiple days by using relatively high-end but still commonly available hardware. In particular, the hardware employed in this analysis included 64-bit machines with 16 GB of RAM and multiple processors allowing for higher-speed parallel processing. Unfortunately, the initial attempts to simulate the full Portland network by using either model were not successful on two levels. The ratio of real time to simulated time was almost 1:1, which meant that the simulation of the 3 p.m. to 7 p.m. time period for a single day required about 4 hours of real time to complete. In the enhanced model environment where travel- ers adjust their path selection based on past experience, at least 35 to 50 simulated days are necessary for network travel times to stabilize around an equilibrium level. Thus, up to 200 hours (about 8 days) would have been needed to reach equilibrium for the Portland area network. This is clearly an impractical amount of time to allocate to such an activity, and particularly so in most day-to-day working environments. Neither DYNASMART-P nor DynusT was able to success- fully complete the required 35 to 50 days of simulated time, as both models crashed after only 1 to 2 days of simulated time. The reason is believed to be the large amount of data that must be carried forward in each iteration, with 1.2 million originat- ing vehicle trips occurring on each simulated day. An attempt was made to overcome these problems by aggre- gating the zones from 2,000 down to only 400 super-zones, and while this action had some beneficial effect, it still did not fully resolve either of the two issues identified above. On the other hand, it was found that both DYNASMART-P and DynusT were able to perform acceptably when modeling the smaller subarea network: a single simulated 4-hour day could be com- pleted with DYNASMART-P in about 30 minutes by using the hardware environment described above, and in only 7 to 10 minutes when the computer processing environment was increased from two to eight cores working in parallel. The DTALite model fared much better. Using the same hardware environment and without any aggregation of zones, a single 4-hour analysis period of simulated time for the entire Portland area network required only about 5 minutes to com- plete when using DTALite. At the time these investigations were ongoing, DTALite was still in some level of development and was not fully comparable to either DYNASMART-P or DynusT, particularly with respect to the manner in which signalized intersection control is modeled. These differences are only temporary and do not have much impact on the over- all time requirement for the simulation. Even so, it was judged that the demonstration would be more robust and informa- tive if consistency could be maintained with the modeling method used for the Dallas–Fort Worth network. Therefore, a two-step process was used to achieve this goal: • DTALite was used to model the entire Portland metropoli- tan area network for a period of 50 simulated days. • The results of the DTALite model were used to create an O-D matrix for the much smaller subarea network, and this became the basis for the DYNASMART-P modeling of the subarea that followed. The following statistics provide a detailed description of the current road network attributes in the test network: • Network data 44 Number of nodes: 858. 44 Number of links: 2,000. 44 Number of O-D demand zones: 208. • Node control type (number of intersections including on- and off-ramps) 44 No control (on- and off-ramps): 689. 44 Four-way Stop: 0. 44 Two-way Stop: 0. 44 Signalized (actuated control): 169. • Traffic control data for signalized intersections 44 Two-phase control intersection: 4. 44 Three-phase control intersection: 96. 44 Four-phase control intersection: 69. 44 Maximum green: 50 seconds. 44 Minimum green: 10 seconds. 44 Amber: 5 seconds. In order to produce a realistic assessment of the entire four-hour simulation period, entering demand levels were adjusted every 15 minutes according to the Portland area demand profile shown in Figure 4.18. Here, the height of each bar represents the ratio of 15-minute demand to the original overall average. User Information Classes The day-to-day learning procedure described earlier in this report is used to model the ways in which drivers choose routes on a daily basis and how they learn from previous travel experiences. In this procedure, drivers have the opportunity to compare their experienced travel times against travel times on alternate paths and then make route changes for the coming days if they find they can save enough time to warrant the

71 switch. For this application, the percentage of drivers who could make a change in any given day was set to 15%, reflect- ing the reality that many people are creatures of habit and unlikely to make route changes right away, if ever. The amount of travel time drivers needed to save before they consider making a route change was set to 5 minutes. Neither of these assumptions is based on actual field data but come from the judgment of the project team members working in conjunc- tion with the Portland Metro’s planning and modeling staff. Finally, the percentage of vehicles assigned to each of three available user classes was as follows: 98% of drivers were assumed to have no access to real-time information about net- work conditions (this user class is referred to as the Unequipped user class); 1% were assumed to have access to pretrip informa- tion only (this user class is referred to as the Pretrip or PT user class); and 1% were assumed to have access to continuous en route information (this user class is referred to as the Equipped or ER user class). Simulation-Based Strategy Evaluation Procedure A straightforward method was developed to test the effective- ness of one or more operational strategies either as stand- alone projects or as alternatives to traditional new construction projects. 1. First, the location of the operational strategy and/or new construction project that was to be tested was identified, and a subarea or network that appropriately surrounds the location was established. 2. Next, geometric, volume, and operational characteristics of each link within the subarea were identified and provided as inputs to the DTA model. Appropriate link, corridor, and/or network performance measures were established for sub- sequent evaluation purposes. 3. The DTA model was then run under three separate regimes in order to effectively use the day-to-day learning process and generate results that could be usefully compared, as shown in Figure 4.19. During the baseline stabilization period (Regime I), the DTA model was simulated for 50 days to achieve equilibrium under a baseline scenario (i.e., without any of the operational strategies or new con- struction projects that are to be evaluated). 4. After the baseline stabilization period was completed, the operational strategies and/or new construction projects to be evaluated were introduced into the network, and the DTA model was run for an additional 30 days of simulated time to allow driver adjustments and to achieve stable conditions under the new scenario. This is referred to as the strategy stabilization. Following immediately on this 30-day period was a simulation of an additional 20 days that formed the basis for the summary results output asso- ciated with the particular strategy being investigated. Adherence to this methodology provided good insights into the effectiveness of the operational strategies and new construction projects being evaluated. 0.71 0.76 0.80 1.09 1.21 1.11 1.06 0.96 0.87 0.90 0.82 0.77 0.00 0.00 0.00 0.10 0.20 0.30 0.40 0.50 0.60 0.70 0.80 0.90 1.00 1.10 1.20 1.30 4:00~4:15 4:15~4:30 4:30~4:45 4:45~5:00 5:00~5:15 5:15~5:30 5:30~5:45 5:45~6:00 6:00~6:15 6:15~6:30 6:30~6:45 6:45~7:00 7:00~7:15 7:15~7:30 Time Period 1 Period 2 Period 3 Period 4 R at io Figure 4.18. Baseline network demand profile.

72 Measures of Effectiveness For the purposes of this demonstration project, a number of performance measures were monitored. Summary results for each performance measure were aggregated on a link, corridor, O-D-pair, and/or network basis, depending on the nature of the performance measure. The performance mea- sures that were monitored in this application included the following: • Peak hour (5 to 6 p.m.) link and corridor volume (vph); • Total travel time (minutes) for links, corridors, and the entire network; • Average travel time (minutes/veh) for links, corridors, and the entire network; • Vehicle miles traveled (VMT) during the peak hour for each corridor and the network; • Average speed (mph) for each corridor and the network; • Density (veh/mi/lane) for each corridor; and • Breakdown frequency for each corridor. The results from the 20 simulated days were also used to compute an average value for each of these performance mea- sures and values representing both the 5th and 95th percentile confidence levels for each performance measure. Identification of Alternative Improvement Strategies Figure 4.20 provides a visual overview of the subarea base- line conditions after completion of the 50-day Regime I period. It is clear that Highway 26 (along the northern edge of the study area) and TV Highway (in the center of the study area) are both east-west facilities with significant amounts of congestion. This is also true in real life: both facilities are heavily used by westbound work-to-home commuters during the evening peak period, and both facilities also provide important regional connectivity services at the same time. On the basis of these facts, it was hypothesized that the following improvement strategies might be effec- tive countermeasures and therefore worthy of additional investigation: 1. Expansion of TV Highway west of Murray Boulevard from the existing five-lane cross section to a new seven-lane cross section through new construction; 2. Addition of two new through lanes in the eastbound and westbound directions on TV Highway west of Murray Networkwide Simulation Results 0% 2% 4% 6% 9% 11% 13% 15% 17% 0 5 10 15 20 25 0 10 20 30 40 50 60 70 80 90 100 Sw itc hi ng % Tr av el Ti m e Days Baseline Travel Time Strategy Travel Time Baseline Switching % Strategy Switching % Figure 4.19. Overview of strategy testing plan under stochastic capacity. Figure 4.20. Overview of Portland subarea baseline conditions.

73 Boulevard, resulting in four through lanes in each direc- tion of travel; 3. Conversion of TV Highway west of Murray Boulevard from the existing five-lane cross section to a new seven- lane cross section through use of narrower lanes and shoulders; 4. Improved signal timing as well as the addition of a new through lane in each direction on Baseline Road (between Murray Boulevard and the City of Hillsboro); 5. Improved signal timing on Baseline Road plus the intro- duction of congestion pricing on TV Highway between Highway 217 and Murray Boulevard, by adding a $1 sur- charge to any trip using any part of TV Highway within these boundaries during the evening peak hour (5 to 6 p.m.); 6. Expansion of the availability of (and access to) pretrip traffic condition reports and information, such that 10% of all drivers take advantage of this opportunity (under baseline conditions, only 1% of all drivers are assumed to consult traffic condition reports/websites before depart- ing on their trip); 7. Expansion in the use of en route information, from 1% under existing conditions to 10%. Evaluation of Alternative Improvement Strategies Each of the seven improvement strategies identified above, in addition to a “no change” baseline scenario, was analyzed. Each analysis began with traffic volumes and routing path conditions set as they existed at the end of the Regime I time period (see Figure 4.19). The improvement strategy was then introduced and an additional 50 days were simulated. Results data were then collected for the last 20 days of this simulation period. Depending on the particular hardware being used, each 50-day simulation required between 6 and 13 hours to com- plete, using 64-bit machines with 16 to 18 GB RAM. More specifically, machines with two parallel processors required 12 to 13 hours to complete a 50-day simulation, whereas machines with four parallel processors required 6 to 7 hours to complete the same task. Large output files are produced from each model run. As an example, approximately 5 million records relating to link per- formance characteristics were produced from each model run. These records were imported into a query (SQL) database and then post processed by using customized but simple routines in order to produce the summary results that follow. Summary of Results Figures 4.21 through 4.25 present summary results across all performance measures for a variety of corridors, for the subarea network as a whole, and for three separate O-D pairs. The indicated vertical lines depict the 95th percentile confi- dence interval for the mean value of the MOE and thus are a measure of the MOE reliability. Examination of these results reveals the following potentially important observations: • Westbound travel time on TV Highway between Highway 217 and Hillsboro is most positively affected by the conges- tion pricing alternative. However, this same effect is not apparent for the section of TV Highway between Murray Boulevard and Hillsboro. Further, the congestion pricing strategy had a net adverse effect on travel times within the network as a whole. • Network travel time performance benefited the most from increasing the percentage of drivers who make use of pre- trip information, although this particular strategy did not significantly improve the travel performance of any of the three east-west corridors that were examined. • The reliability of travel times on TV Highway between High- way 217 and Murray, as defined by the difference between the 5th and 95th percentile confidence levels, improved the most with the construction of an additional through lane in each direction on TV Highway; the provision of an addi- tional lane and improved signal timing on Baseline Road; and increased usage of pretrip information. In other words, improvements in travel time reliability were as great with some low-cost strategies as was achieved with the construc- tion of an additional lane, even when average travel time was not appreciably affected. • All tested strategies resulted in a fairly significant reduction in average corridor density for each of the three corridors that were examined. However, this did not always translate into a corresponding reduction in average travel time. These types of findings, which are an outcome of the enhanced network operational analysis procedures described in this report, provide a more detailed and complete assess- ment of the effectiveness of operational strategies at a network level. It is a significant step forward from the level of informa- tion produced by traditional transportation planning model- ing and analysis tools. The types of findings described above are useful for not only transportation professionals but also decision makers who have responsibility for transportation investment decisions in the subarea boundaries. The analysis reveals some fundamental take-away points for transportation professionals who undertake analyses of this type: 1. Multiple performance measures must be considered in order to obtain the most complete assessment of a par- ticular operational improvement strategy. Levels of ser- vice and volume/capacity ratios continue to be important, (text continues on page 78)

74 27.5 28 28.5 29 29.5 30 30.5 31 1 Tr a ve lT im e (m inu tes ) Scenarios TV Hwy Travel Times (WB - 217 to Hillsboro) 0 1,000 2,000 3,000 4,000 5,000 6,000 7,000 8,000 1 VM T Scenarios TV Hwy VMT (WB - 217 to Hillsboro) 0 2 4 6 8 10 12 14 16 18 1 D en si ty (ve h/m i/ln ) Scenarios TV Hwy Density (WB - 217 to Hillsboro) 18.5 19.0 19.5 20.0 20.5 21.0 21.5 1 Sp ee d (m ph ) Scenarios TV Hwy Speed (WB - 217 to Hillsboro) 0 50 100 150 200 250 300 350 1 # of C yc le F ai lu re s Scenarios TV Hwy Cycle Failures (WB - 217 to Hillsboro) Cong. Pricing + imp. Sig. New const.: +2 lanes Narrow lanes Baseline Rd imp. Signal Baseline Rd imp sig +1 Lane 10% Enroute Info 10% Pre-Trip Information Baseline Rd Improve. New const.: +1 Lane Existing Condition Figure 4.21. Performance characteristics of alternative improvement strategies on TV Highway (between Highway 217 and Hillsboro).

75 Cong. Pricing + imp. Sig. New const.: +2 lanes Narrow lanes Baseline Rd imp. Signal Baseline Rd imp sig +1 Lane 10% Enroute Info 10% Pre-Trip Information Baseline Rd Improve. New const.: +1 Lane Existing Condition 21 21.2 21.4 21.6 21.8 22 22.2 22.4 22.6 22.8 TV Hwy Travel Time (WB - Murray to Hillsboro) 1 Tr a ve lT im e (m inu tes ) Scenarios 0 1,000 2,000 3,000 4,000 5,000 6,000 7,000 TV Hwy VMT (WB - Murray to Hillsboro) 1 VM T Scenarios 0 2 4 6 8 10 12 14 16 TV Hwy Density (WB - Murray to Hillsboro) 1 Scenarios 21.6 21.8 22.0 22.2 22.4 22.6 22.8 23.0 23.2 TV Hwy Speed (WB - Murray to Hillsboro) 1 Sp ee d D en si ty (v eh /m i/ln ) (m ph ) Scenarios 0 50 100 150 200 250 TV Hwy Cycle Failures (WB - Murray to Hillsboro) 1 # of C yc le F ai lu re s Scenarios Figure 4.22. Performance characteristics of alternative improvement strategies on TV Highway (between Murray Boulevard and Hillsboro).

76 0 5 10 15 20 25 30 Baseline Rd Travel Time 1 Tr a ve lT im e (m inu tes ) Scenarios 0 1,000 2,000 3,000 4,000 5,000 6,000 7,000 8,000 9,000 Baseline Rd VMT 1 VM T Scenarios 0 2 4 6 8 10 12 14 16 18 20 Baseline Rd Density D en si ty (ve h/m i/ln ) 1 Scenarios 0 5 10 15 20 25 30 Baseline Rd Speed 1 Sp ee d (m ph ) Scenarios 0 20 40 60 80 100 120 140 160 180 200 Baseline Rd Cycle Failures 1 # o fC yc le Fa ilu re s Scenarios Cong. Pricing + imp. Sig. New const.: +2 lanes Narrow lanes Baseline Rd imp. Signal Baseline Rd imp sig +1 Lane 10% Enroute Info 10% Pre-Trip Information Baseline Rd Improve. New const.: +1 Lane Existing Condition Figure 4.23. Performance characteristics of alternative improvement strategies on Baseline Road (between Murray Boulevard and Hillsboro).

77 Imp Signal + cong Pricing New const Narrow lanes Imp Signal Imp sign + new lanes Pretrip Baseline (Rd) Imp Exist cond Exist (2 in TV) 9.20 9.70 10.20 10.70 11.20 11.70 12.20 Tr av el Tim e (m inu te s) Scenarios Network Travel Time Figure 4.24. Networkwide travel time characteristics of alternative improvement strategies. Imp Signal + cong Pricing New const Narrow lanes Imp Signal Imp sign + new lanes Pretrip Baseline (Rd) Imp Exist cond Exist (2 in TV) 0.00 1.00 2.00 3.00 4.00 5.00 6.00 7.00 8.00 9.00 1 Tr av el Tim e (m inu te s) Scenarios Critical OD 1 12.50 13.00 13.50 14.00 14.50 15.00 15.50 16.00 16.50 17.00 1 Tr av el Tim e (m inu te s) Scenarios Critical OD 2 0.00 1.00 2.00 3.00 4.00 5.00 6.00 1 Tr av el Tim e (m inu te s) Scenarios Critical OD 3 Critical OD 1: A-B Critical OD 2: C-B Critical OD 3: B-A Figure 4.25. Performance characteristics for three origin–destination (O-D) pairs.

78 of course, but these by themselves do not provide a complete or even adequate assessment of an operational improvement strategy, particularly in situations where a range of improvement strategies is being tested. 2. Performance measures must be selected to reflect link, cor- ridor, and network characteristics. Alternative improve- ment strategies can have markedly differently effects at the link, corridor, and network levels, both individually and also relative to one another. 3. Evaluations of operational strategies at the network level must consider impacts on driver’s route choice. As a result of the modeling enhancements made as part of this proj- ect, the Portland subarea exercise demonstrated the effects that operational strategies have on route choice. In some cases, operational strategies such as added lanes result in capacity increases that attracted vehicles to a particular route, whereas other strategies such as traveler informa- tion shifted demand from congested facilities to routes with available capacity. Understanding the relationship that capacity enhancements have on demand and vice versa is critical for congested networks. 4. A representative cross section of corridors and O-D pairs should be evaluated. Past analyses have typically been lim- ited to the particular links and/or corridors within which the improvement strategy is implemented. However, the results of this demonstration application show that the per- formance characteristics of other corridors and O-D pairs are also likely to be affected and should therefore be taken into account. 5. The effectiveness of a particular operational improvement strategy can only be evaluated within the context of the network environment in which it will be implemented. It is not possible to accurately estimate the capacity-enhanc- ing effects of a particular strategy through something as simple as a lookup table; so many other factors affect the effectiveness of an operational improvement strategy that it must be assessed in the context of the network within which it is situated. Travel time reliability is an especially important perfor- mance measure to consider within a congested or oversaturated network. As demand outstrips supply and congestion levels increase, it becomes more and more difficult to show significant improvements in average travel time through the implementa- tion of one or more operational improvement strategies. But quite often the reliability associated with the average travel time will improve with these operational improvements, even when average travel times remain virtually unchanged. This is a very important benefit because it has the effect of giving more discretionary time back to the driving public, just as an abso- lute reduction in average travel time would have done. It is also a benefit that has previously gone unnoticed, unmeasured, and/or unreported in analyses of this type. Summary and Conclusions The Portland subarea network represents a good venue for demonstrating the steps involved in applying the methodolog- ical enhancements developed in this project to a real-world environment. The demonstration described in this chapter has also documented the value and effectiveness of this new analy- sis tool in transforming the way alternatives analyses are con- ducted and, ultimately, the way important transportation investment decisions are made. Important insights have been achieved that will positively affect the nature, scope, and scale of performance measures used to judge different investment strategies. reference 1. Mahmassani, H., and H. Sbayti. DYNASMART-P User’s Guide Version 1.2. Maryland Transportation Initiative, 2005. (continued from page 73)