Objective 1: Performance Measures for a Signalized Arterial System

Transcription

1 NCHRP Project 3-79a Working Paper No. O1-2 Objective 1: Performance Measures for a Signalized Arterial System Prepared for: National Cooperative Highway Research Program Transportation Research Board National Research Council Transportation Research Board NAS-NRC LIMITED USE DOCUMENT This report is furnished only for review by members of the NCHRP project panel and is regarded as fully privileged. Dissemination of information included herein must be approved by the NCHRP. Prepared by: Chris Day and Darcy Bullock June 1, 29

2 Objective 1: Performance Measures for a Signalized Arterial System Contents Abstract... 2 Introduction... 3 Framework for Discussion... 6 Local Intersection Performance Measures... 6 Arterial Phase Volume and Green Time... 6 Degree of Intersection Saturation... 6 Volume-to-Capacity Ratio and Split Failures... 8 Arrival Type... 1 Examining Volumes for Analyzing Time-of-Day Plan Breakpoints Estimated Delay Longitudinal Data Moving Toward Active Arterial Traffic Management Conclusion References... 17

3 June 1, 29 Abstract Most urban traffic control systems go through a rather rigorous design phase based upon a set of fixed design volumes that do not capture the stochastic variation in traffic due to weather, incidents, special events, and shifting demand patterns. Once these systems are built, their operation is relatively open loop, with public feedback (complaints) often the primary feedback for assessing operations and initiating changes. This paper extends the concept introduced in the first working paper to define procedures and case studies that illustrate how fundamental traffic engineering concepts can be integrated with traffic signal system detection and controller status information to provide system performance measures. These performance measures characterize the operation of a traffic signal system and provide a structure procedure for prioritizing operation improvement opportunities. A consistent set of performance measures used across an agency will impact traffic signal operations at two distinct levels: At the district/sub-district it would provide a continuing updated list of hot spots requiring attention. These hot spots would be much more reliably defined in terms of location and time then telephone reports. At the state level, it would provide the agency with quantitative data for prioritizing resources across districts. This data would be tabulated on a consistent basis from district to district. Several example performance measure graphics are provided to illustrate how these tools can be used for prioritizing operations decisions. 2

4 June 1, 29 Introduction Signalized arterials represent a significant component of the highway transportation network in the United States. The task of maintaining appropriate signal timing plans is of paramount importance, but for many agencies there is a scarcity of engineering resources needed to monitor and update those plans. While timing plans are often designed to a high degree of rigor, these plans often do not take off-peak and weekend conditions into consideration, nor do they consider variation due to shifting patterns in demand. In many cases, signal timings are updated in response to user complaints, or are otherwise maintained on very long timelines. In both cases, the information needed to respond to changing conditions takes considerable effort to gather, and in many cases may not be able to capture specific aspects of operation that lead to problems. In the previous working paper, a number of performance measures were described that provided information on conditions at a single signalized intersection, taking into consideration not only the traffic volumes served, but also the distribution of capacity among the competing phases and the quality of arterial progression. In this paper, we expand the scope to a scale at which agencies can assess the performance of a system of signals. An example of how system capacity problems may be identified at a high level is shown in Figure 1. Here, the network of interest is the US interstate system and the performance measure is the total number of hours of truck delay caused by bottlenecks at interchanges (1). The map (Figure 1a) shows the geographic location of areas where the system has problems. That areas such as Chicago, New York, and Los Angeles have a high number of congested interchanges is no surprise, but that there should be moderate clustering around Columbus, Ohio or Phoenix, Arizona is a detail less likely to be known on a national level without performance measure data. The magnitude of delay caused by each bottleneck is shown comparatively in Figure 1b by a sorted bar graph. This distribution shows that there are a handful of bottlenecks where delay is considerably more severe than the rest of the system, information that would be instrumental in prioritizing interchanges for receiving funding for improvements, if reduction of truck delay is one of the variables for allocating such funds. Agencies that manage signal systems operate at the level of municipalities, counties, and regions of a state, as illustrated by Figure 2. In Indiana, for example, the task of managing state highways is divided into six districts, which are further divided into subdistricts that consist of perhaps two or three counties. A single county may have several dozen or more signals; the state agency manages approximately five thousand signals overall. To maintain these systems, there are only a handful of engineers in each district and perhaps a dozen or fewer in the statewide central office, many of whom are tasked with other responsibilities besides maintaining traffic signals. As shown by the county map, the physical dispersion of these systems and the mixture of arterial systems and isolated signals would tend to make more expensive adaptive control systems infeasible. The creation of a performance monitoring module capable of reporting the operating status at each location would be a significant first step in providing better service at signals throughout the system. Because there is currently no internal reporting system, problems in an arterial system come to light by external means to the system. One major source of this information is anecdotal evidence of problems obtained from user feedback (i.e., 3

5 June 1, 29 complaints).users are likely to report that a particular arterial is experiencing problems, but these reports generally do not give precise details on the time or extent of the problem. A consistent set of performance measures used across an agency would impact traffic signal operations at two distinct levels: At the district/sub-district it would provide a continuing updated list of hot spots requiring attention. These hot spots would be much more reliably defined in terms of location and time then telephone reports. At the state level, it would provide the agency with quantitative data for prioritizing resources across districts. This data would be tabulated on a consistent basis from district to district. Performance anomalies would likely be substantiated by further investigation such as a travel time study (2). An example of the results of such a study are shown in Figure 3; this shows travel times obtained for eastbound movement on SR 26 in Lafayette, Indiana. A considerable amount of travel time is added to this commute at various signals along the way, especially at the intersection of US 52, another signalized arterial, that took two cycles to clear. Although this analysis adds some precision to a more vague report that there is a high amount of delay going east on SR 26, this information has not given us detailed information about the condition of the competing phases at each intersection. Furthermore, the GPS run represents only one particular time of day and it is unclear whether the conditions are recurring or nonrecurring. Data collected internal to the signal system can be used to reveal added dimensions of the problem and help understand the underlying causes. Figure 4 shows a plot of the percentage of vehicles arriving during green at SR 37 and Pleasant in Noblesville, IN (Intersection 12 in Figure 5). The bottom part of the graph is a progression diagram, which was described in detail in the previous working paper. This diagram indicates the relationship of each vehicle arrival in the system to the start and end of green for the green phase for a coordinated through movement. The x-axis of the graph represents the beginning of the cycle (time ); the green lines show the start of the coordinated green phase, and the red lines show the end of the phase (also the end of the cycle). Each dot shows an actuation of the advance detector. The clustering of dots between the red and green lines indicates a majority of arrivals on green, which is representative of good progression. However, there are certain times of day, such as 13: 15: (the circled region in Figure 4), where there are many vehicles arriving on red. In this example, the arrival of the coordinated platoon during the red phase suggests that the offset should be adjusted. Figure 5 outlines a strategy for managing closed loop systems that explicitly incorporates performance measures in the management of traffic signal systems. In such a scheme, data is continuously collected throughout the system, and performance measures are tabulated and archived in a central database. At regular intervals, the performance of the system is reviewed. Subject to the agency objectives and constraints set by decision makers, the decision is made whether to intervene and make 4

6 June 1, 29 adjustments to signal plans in parts of the system. In deciding whether to intervene, the questions that the traffic engineer asks are oriented toward capacity and progression: With regards to capacity: 1. Which intersections have the most capacity deficiencies? 2. Are these deficiencies recurring or non-recurring? 3. At intersections with deficiencies, is there sufficient unused capacity at certain time periods of the day to mitigate deficiencies by reallocating green times? 4. At what periods of the day do opportunities for mitigation occur? 5. During those time periods, which phases should be allocated additional capacity and which phases could perform acceptably with a reduced capacity allocation? With regards to progression: 1. Which intersections have the most progression deficiencies? 2. Are these deficiencies recurring or non-recurring? 3. Can the problem be solved with offset adjustment, use of lead/lag left turn phases, or pattern change? 4. At what periods of the day do opportunities for mitigation occur? 5. During those time periods, will adjustments to the plan improve arterial progression without causing capacity problems and without disturbing progression in the other direction? Currently, none of the major manufacturers of signal controllers provide an internal module for collecting cycle-by-cycle performance measures. Historically, bandwidth has been the limiting factor for communications in closed-loop systems. However, as the cost per unit of bandwidth continues to decrease and the market for ITS technologies increases with initiatives such as IntelliDrive (3), there can be little doubt that the communication infrastructure will improve. Newer models of signal controllers from major manufacturers (Econolite ASC/3, Siemens Eagle ATC NX, Peek ATC TS/2, Naztec Model 98, McCain 27L) feature Ethernet connection ports. Table 1 gives a rough estimate of data requirements for monitoring an agency network of 2 signals. Table 1a shows the needed storage for recording all of the raw events (3) needed to generate performance measures as described in the previous working paper. We assume that six bytes are needed per event, using 4 bytes to record a timestamp and 2 bytes to include the event type and a parameter. We assume that time would be stored as the 32-bit UNIX timestamp. It might be possible to use a smaller timestamp and achieve greater economy of bandwidth. Given 1, cycles taking place at an eight phase intersection serving 5, vehicles, an estimated 184,1 events are needed to describe the state of the signal throughout the day. Assuming that 6 bytes are needed to store each event, approximately 1.1 million bytes are needed to store 24 hours of signal data. Archiving this information for a network of 2 signals over one year would require 8.6 billion bytes. Far fewer bytes are needed to store the performance measures alone. Table 1b shows the needed storage for a database of basic performance measures. From these measures, higher level metrics can be derived; for example, the volume-to-capacity ratio can be calculated from a database view taking into consideration vehicle counts, 5

7 June 1, 29 green times, and cycle lengths. For 1, cycles having eight phases, approximately 1, bytes are needed for a 24 hour period, which scales up to 7.3 billion bytes to archive information for a 2 signal network over a year. If the performance measures are calculated at the controller, the raw events do not need to be transferred and the amount of bandwidth required is reduced by more than a factor of 1. Framework for Discussion This paper uses SR 37 in Noblesville, Indiana as an example signalized arterial for discussing system performance measures. A map of the system is shown in Figure 6. Intersection 11 (SR 32 and SR 37) is the master intersection; this was the example intersection used in the previous working paper. The data presented in the figures was collected on Wednesday, April 8, 29. Local Intersection Performance Measures The previous working paper advocated the concept of estimating a variety of traffic engineering performance measures on a cycle by cycle basis to identify problems. Those concepts were developed for a single intersection. Subsequent sections of this paper extend those concepts and explain how they can be aggregated and analyzed on a system basis. Arterial Phase Volume and Green Time The most basic measurements for signal operation are vehicle counts and capacity allocation. While counts are generally known from engineering studies used to create signal timings, and capacity allocation is largely determined by splits, it is informative to be able to track these numbers to validate that the design counts reflect current served volumes, and that the signal timings are reasonable. Figure 7a shows a map of total 24 hour served volumes by phase for all 32 phases in the four-intersection system. Not surprisingly, the northbound and southbound phases on the arterial have the largest numbers of vehicles served. Although these vehicle counts may be compared against expected traffic patterns to determine whether demand patterns have shifted or not, they provide little insight into how the system is performing. The total green time provided at each intersection is shown in Figure 7b. This shows the response of the signal controller to conditions at each intersection. On first glance, these numbers seem to follow the same trends as the volumes in Figure 7a; this would be reasonable. Although a few anomalies may be identified on inspection, this aggregated green time provides little insight into how well the capacity is matched to demand throughout the day. Degree of Intersection Saturation The degree of intersection saturation (X C ) is a Highway Capacity Manual (5) metric that describes the overall utilization of the intersection. The formula is: 6

8 June 1, 29 V C XC =, Equation 1 i s C L ci where: C = cycle length (s), L = lost time (s), and (v/s) ci =the summation over critical phases ci of the ratio of volume (V) to saturation flow rate (s). For a typical dual-ring eight phase controller, this equation simplifies to: v v56 v34 v C = max 12, + max s s s,, Equation 2 s C L X C 78 Where, for example, v 12 /s = v 1 /s + v 2 /s. Calculation of this performance measure at a local intersection was described in more detail the previous white paper. Twenty-four hour plots of X C are shown in Figure 8, showing the utilization of capacity at the four arterial intersections. This figure shows similar peaking trends for all of the intersections, with AM peaks occurring at 9: and PM peaks occurring at 18:. Intersections 11, 12, and 14 utilize most of their capacity for much of the day, whereas intersection 13 appears to have a considerable amount of slack. The purpose of such plots are to indicate to the traffic engineer at what times of day there is spare capacity within the existing cycle length; at these times there are opportunities to adjust splits to serve phases that are experiencing capacity deficiencies. The cutoff point is a value to be determined by agency policy that essentially tells the traffic engineer when the effort to retime the system succumbs to the law of diminishing returns. As discussed in the previous white paper, as X C approaches 1, there are fewer seconds of underutilized green time to be redistributed. The X C plot also shows times of day when such opportunities do not exist, and where all capacity is being used (i.e., when X C approaches or exceeds 1). Not surprisingly, more capacity is utilized during the peak periods. However, these plots clearly show that some intersections have a significant amount of spare capacity even during the peak hours that could potentially be rebalanced. The distributions of X C at each intersection may be visualized by sorting the values from greatest to least. Figure 9a shows such a plot; each value of X C in the graph corresponds to a point in Figure 8. This plot shows the overall utilization of each intersection throughout the day. The region of concern, being the cycles where X C exceeds.75, is magnified in Figure 9b. Intersection 11 reports over 1 cycles where X C exceeds the cutoff value, indicating that split adjustments alone might not be the appropriate approach to operational issues at this intersection, at least during a substantial portion of the day. In contrast, intersection 13 does not ever enter this region, indicating that split adjustments would be an appropriate response. 7

9 June 1, 29 Volume-to-Capacity Ratio and Split Failures The volume-to-capacity (V/C) ratio of phase i is calculated by (5): v V V C X = = i i i =, c i g s i i g i si C i Equation 3 Where: X i = the v/c ratio for phase i, V i = the flow rate for phase i (veh/h), s i = the saturation flow rate for phase i (veh/h), g i = the effective green time for phase i (s), and C = cycle length (s). This measure can be used to determine whether a phase failure has taken place. A split failure can be defined as an occurrence when there is not enough green time to serve the demand. As X i ratio increases, it becomes more likely that a split failure occurs. For the sake of expediency, we select X i = 1 as the threshold for determining when a split failure takes place 1. As mentioned in the previous white paper, saturation flow rate is an important characteristic that strongly affects the calculation of X i, and must accurately reflect the driver behavior at the intersection. The V/C ratio is a phase-based measure; between the four intersections in the arterial system, there are 32 phases for which X i may be calculated on a cycle-by-cycle basis. A plot with 32 series would not be helpful for visualizing this data for the arterial. However, it is possible to combine and plot the X i values for all eight phases into one sorted series for each intersection, similar to the approach for for X C in the previous section. An example of such a plot is shown in Figure 1a. The top 1 X i values are shown in Figure 1b. While this plot loses all of the phase-specific information, it illustrates the general characteristics of capacity utilization, and it shows which intersections experience the greatest number of split failures. In this case, it is clear that intersection 12 experiences the highest number of capacity problems; approximately 25 split failures occur during the 72 or so phase instances; Intersection 13, in the meanwhile, has relatively few split failures. It may be that one particular phase is responsible for those few failures that occur. Once problem intersections are identified, it is then possible to address the individual phases by plotting a limited number of sorted X i values for the intersections of interest. Here, we have broken out individual phases for all four intersections in the system. Figure 11 shows plots of eight phases at intersections 11 (Figure 11a) and 12 (Figure 11b), while Figure 12 shows intersections 13 (Figure 12a) and 14 (Figure 12b). These distributions describe the magnitude of the utilization of each phase. The number of points above the X i = 1. line represent the number of split failures. These phases would be candidates for receiving additional green time, while 1 We intend to study the validity of this cutoff point and investigate cycle-by-cycle fluctuations of saturation flow rate in future tasks. 8

10 June 1, 29 the phases with lower distributions would be candidates for giving up some green time. In these plots, the following individual phases stand out as having problems: Phases 3, 4, and 5 at intersection 11 Phase 1 at intersection 12 Phase 3 at intersection 13 Phase 4 at intersection 14. Also of note is the width of the series. Phase 3 at intersection 12, for example, has a number of points above the 1. line, but the narrow width of the distribution shows that this phase is rarely served. It might not be an appropriate recipient of additional green time. This may be compared to phase 2 at intersection 11, which is called during every cycle. Interrogating the data to view only v/c ratios from a particular time of day increases the usefulness of such plots in making decisions on splits for a TOD plan period. A more detailed analysis of local intersection phase utilization based from similar plots of cumulative v/c ratio is included in Day et al. (6). Split failures, estimated as phase instances where X i > 1., can be identified using a map view as illustrated in Figure 13. Figure 13a shows the total number of split failures for each phase at each intersection in the system, while in Figure 13b these numbers are represented as a percentage of the total number of cycles in which the phases failed. The numbers in parentheses show the number of phase instances that failed, which excludes cycles in which the phase was not actuated. The zeros indicate phases that do not indicate problems in serving their demand. The worst failure rate was for the northbound left turn at intersection 12. This phase reported a X i > 1 for 11.1% for all cycles in the day, or 15.4% for all cycles in which the phase was served. Other phases with problems include the northbound left turn at intersection 11 and the eastbound through movement at intersections 12 and 14. Intersection 12 has slightly more split failures than intersection 11 (252 vs. 236), which is rather unexpected since intersection 11 had a slightly higher X C distribution (Figure 9). The reason why this occurs is that X C is a more generic metric that does not use green time and only considers volumes. Also, the split failures are distributed across cycles differently at the two intersections. This illustrates why it is helpful to examine individual phase performance in addition to overall intersection performance. Another way to classify split failures is to look at consecutive split failures, which is defined as when the same phase fails in two consecutive cycles. Figure 14 illustrates this concept graphically. Consecutive split failures indicate that queues might be taking many cycles to clear. This is a potential indicator for storage lane overflow, and possibly for spillbacks if the phase controls traffic feeding from a short segment. Figure 15a shows a map of the 32 phases on SR 37, showing the total number of consecutive split failures that occurred throughout a 24-hour period. The northbound left turns at intersection 11 and intersection 12 and the eastbound through movement at intersection 14 are the most problematic phases. Figure 15b shows the maximum number of times in a row that each phase failed. The most severe failure pattern occurred for the eastbound through movement at intersection 12; at one point 9

11 June 1, 29 during the day, this phase failed five times in five consecutive cycles. This phase would be a candidate for receiving additional green time in a split adjustment. The possibility of downstream blockages should be taken into consideration when characterizing phase utilization from X i values. In a downstream blockage scenario, the affected phase is shown the green indication, but vehicles are unable to leave the approach because there is nowhere to go. In this case, the vehicle count would be low or even zero, whereas occupancy would be high. Methods to detect this situation rely on investigating the stop bar detector occupancy (7). When high occupancy during green coincides with low vehicle counts, a downstream blockage is suspected. Arrival Type Arrival type is a Highway Capacity Manual (5) performance measure that describes the quality of progression. Arrival types are defined from a quantity called the platoon ratio, R p, which is given by: POG C R = = POG i p, i i, gi gi C Equation 4 Where: R p,i = the platoon ratio for phase i C = cycle length (s), g i = green time for phase i (s), and POG i = the proportion of vehicles arriving on green. Table 2 shows the HCM definition of qualitative arrival types based on R p. An arrival type of 6 represents excellent progression, while 1 represents very poor progression quality. In this study, we have interpolated between categories to retain the precision of the R p calculation while making use of the qualitative arrival type scale. Figure 16 shows maps of arrival type during the AM (Figure 16a) and PM (Figure 16b) peak hours along SR 37. Because this arterial is located to the north of Indianapolis, the primary movement in the AM peak is southbound, while in the PM peak the northbound takes priority. The southbound movement at intersection 11 is about 1 mile from the upstream intersection; the arrival type for that movement reflects random arrivals. Generally, arrival types reflect favorable progression for the priority movement during the peak period. We would question whether the appropriate offset is being used at intersection 12 during the AM peak, which seems to be serving northbound vehicles better than southbound, contrary to expectations. There might be an opportunity to improve southbound coordination at that time of day. Northbound progression in the PM peak is generally favorable; intersection 13 in particular is very well coordinated. Twenty-four hour plots of arrival type are shown in Figure 17 for the northbound (Figure 17a) and southbound (Figure 17b) coordinated through phases. The northbound movement is well coordinated during the PM peak hour, as shown by arrival 1

12 June 1, 29 types in the 4 5 range between 15: and 19:. With the exception of intersection 11 around noon, the northbound movement seems to suffer between 9: and 15:. As for the southbound movement, there are few strong trends in the data. Intersections 13 and 14 are well coordinated throughout most of the day. Intersection 11 reports random arrivals over 24 hours, which is expected since arrivals here are random. Southbound progression at intersection 12 seems to suffer after noon; this might be inescapable since the northbound movement is favored at that time, but there may be room for improvement. As was done earlier with X C and X i, a plot of sorted arrival type values can be used to assess the quality of progression for multiple phases in the system. It does not make sense in this case to compare northbound and southbound arrival types on the same plot, so we do not face the problem of having large numbers of series as for X i. Also, because of the different trends in the AM and PM peaks, it is advisable to create separate plots for each peak period. Figure 18 shows sorted arrival type plots for the northbound phases during the AM (Figure 18a) and PM (Figure 18b) peaks, while Figure 19a and Figure 19b show the respective plots for the southbound phases. There are relatively few points in these plots where there were arrival types of less than 2, indicating poor progression; most of these points are observed in the northbound direction during the AM peak (Figure 18a) and in the southbound direction during the PM peak (Figure 19b). This is not surprising, since these represent the lower priority movements during those peak periods. The favored movements seem to perform well; the northbound PM peak movement (Figure 18b) has many points where arrival type is greater than or equal to 4, indicating good quality progression. There appears to be room for improvement in the southbound AM peak movement at intersection 12 (Figure 19a), considering that the northbound movement performs much better during the same time (Figure 18a). Examining Volumes for Analyzing Time-of-Day Plan Breakpoints When considering changes to a signal timing plan, one set of considerations are the points at which to set breakpoints in the time of day (TOD) plan. Plots of the northbound and southbound arterial volumes for the four intersections are shown in Figure 2a and Figure 2b respectively. The vertical lines represent the actual TOD breakpoints currently in use on the arterial. Figure 21 shows the same plots with the points removed for clarity. These plots show that the AM and PM peak volumes are captured reasonably well by the current TOD plan. Southbound volumes seem to begin rising somewhat earlier than 6:, so there may be some benefit in starting the AM peak interval earlier. It may not be necessary to use three separate TOD plans for 9: 15:, since volumes do not vary strongly during this time period. One plan might be sufficient, and would eliminate two potential sources of coordination problems due to plan transitions. 11

13 June 1, 29 Estimated Delay Control delay may be estimated using techniques described in the previous white paper. In this paper, we will use the Input-Output (IO) method developed in NCHRP project 3-79 (8, 9) as a means of estimating delay. Figure 22 shows plots of IO delay for the four northbound (Figure 22a) and southbound (Figure 22b) movements on SR 37. The four lines in this plot represent 2- point moving averages, which approximates the central tendency of the delay throughout the day. Intersection 12 seems to have some unusually high delay from 13: 15: in both northbound and southbound directions. The southbound movement at intersection 11 has rather high delay from 6: 22:, which is not unexpected considering the rather poor arrival types because of random arrivals for this movement. Figure 23 shows the average values of IO delay for the arterial northbound and southbound movements along SR 37 during the AM peak (Figure 23a) and PM peak (Figure 23b). As might be anticipated from the 24 hour plots in Figure 22, the southbound movement at intersection 11 has the highest delay during both the AM and PM peaks. The northbound movement at intersection 14 in the AM peak and the southbound movement at intersection 12 during the PM peak also have relatively high delay. Based on this information, it is possible to construct virtual probe vehicle trajectories (1), as shown in Figure 24, where northbound (Figure 24a) and southbound (Figure 24b) trajectories are shown during the AM and PM peak periods. At each intersection, the distance from the free-flow line to the virtual trajectory corresponds to the amount of delay incurred on average during the time period. For example, if we follow the southbound PM peak travel line in Figure 24b, at intersection 11 it incurs 45.4 seconds of delay; at 12, 23.6 seconds; then 12.8 and 18.3 seconds at intersections 13 and 14 respectively. The amount of delay contributed by intersection 12 is shown in Figure 24b. In the plot, delay is represented as the virtual probe vehicle coming to a halt at each intersection for this amount of time. This is an approximation (compare to Figure 3), and is not necessarily a reflection of a typical vehicle trajectory (i.e., there may be little or no actual stopped delay). There appears to be more delay during the PM peak for both directions, which is characteristic of more activity due to the presence of shopping and restaurants along the arterial. In the southbound direction, a substantial amount of delay is incurred at intersection 11, which is not coordinated. Southbound delay is substantially higher during the PM peak, which is expected since coordination favors the northbound direction at that time of day. There is little difference between northbound delay in the AM and PM peak periods. Longitudinal Data Another aspect of signal operation is the recurrence of operating problems. Major anomalies of demand due to an incident or special event are well known to cause issues in operation. However, less drastic shifts in demand may also be recurring or nonrecurring. A longitudinal analysis, which compares data over a continuous time period, reveals whether such patterns are sustained or transient. The purpose of carrying out this analysis would be to determine whether a timing plan should be 12

14 June 1, 29 changed. However, if appropriate data is gathered during special events it may be possible to plan ahead to create future plans to serve traffic during known similar events in the future. To illustrate the impact of an event on signal operation, we use the example of the last two full weeks in December 28, which shows the impact of the Christmas holiday on traffic patterns. Figure 25 shows a twelve-day plot of X C at intersection 11 (SR 37/32) in Noblesville, Indiana. The impact of the holiday (Thursday, Dec. 25) is quite clear in this figure. If we overlay the week of the holiday on top of the previous week, as in Figure 26, the differences become more apparent. In addition to being far less utilized on the holiday, traffic does not make much of an appearance until much later in the day than normal. The pattern during the rest of the week is also considerably different than the previous week. While for the week of Dec. 15, the AM and PM peaks can be identified in the X C plots, this characteristic is less apparent in the plot for the week of Dec. 22. In particular, Wednesday, Dec. 24 shows a strong midday peak that likely reflects increased travel and retail activity. As for Friday, Dec. 26, a day that many people likely did not work, the traffic pattern was more similar to a weekend. Figure 27 shows the IO delay for the northbound movement at the intersection 11 over the same time period. Following the same trends as X C, the amount of delay is considerably reduced during the holiday (Thursday, Dec. 25), while delay on the preceding and following days appears to follow different patterns. Moving Toward Active Arterial Traffic Management Ultimately, the purpose of obtaining performance measures on a signalized arterial is to manage the system. The feedback loop (Figure 5) of system monitoring and evaluation, currently managed on years-long schedules, should be condensed into a much shorter timeframe once the process of data collection is automated. It is possible to envision a system that is able to alert the traffic engineer to potential problems. As an example of the use of real time performance measures to actively manage an arterial signal, we return to the previous example of arrival type at intersection 11 (SR 37 and SR 32 in Noblesville, Indiana). Recently, concerns about progression along this arterial led to an adjustment of timing plans at the four signals. As can be seen in Figure 17, the arrival types for the northbound movement are generally poor during the 9: 15: time period. Figure 28 shows a plot of the arrival type for the northbound movement before (Feb. 27, 29) and after (Mar. 2, 29) the retiming effort. The blue before line is similar to the curve in Figure 17, which had rather poor progression during the midday off-peak periods (9: 11:, 13: 15:). The retiming of the signal eliminated these troughs, while northbound progression appeared to suffer during the other time periods. While the changes to the timing plan benefited the northbound movement during the off-peak periods, it is questionable whether the changes during the PM peak should have been made. Figure 29 shows a plot of input-output delay from the same before and after periods. Delay was substantially reduced during the 9: 11: time period. Although delay did increase during the AM peak (the time of day in which the southbound movement is favored), for the rest of the day it remained very similar to the case before retiming. 13

15 June 1, 29 To understand the reason for these changes, progression diagrams were generated for this movement before (Figure 3) and after (Figure 31) the timing plan. As can be seen in Figure 3, the poor arrival types during the off-peak periods might be attributable to two observable occurrences: Clusters of vehicles (circles A, B) arrive during the phase red interval. These appear to be vehicles that turn in from the side street at the upstream intersection. The coordinated platoons (circles C, D) seem to begin appearing just before the start of green. In the progression diagram for operations after retiming (Figure 31), the coordinated platoon in the off-peak periods appears better covered by the green window (Figure 31, circles A, B). There are fewer vehicles arriving in red; the clusters in the red phase visible before retiming (Figure 3, circles A,B) do not appear in Figure 31. During the PM peak period, the coordinated platoon appears very well captured in both before (Figure 3, circle E) and after (Figure 31, circle C) conditions. The lower arrival type (Figure 28) can be explained by the fact that there are more secondary vehicles arriving during red. The reason for this is not clear, but might reflect a exogenous change in activity patterns between February 27 and March 2. Illustration of System Operating Conditions During A Controller Failure The impact of operational problems at one intersection on the other intersections in the system is illustrated in another example at SR 37. During the week of April 12, 29, the controller at intersection 12 failed, and was replaced by a backup unit that did not have the current intersection settings programmed. For 48 hours, this controller operated that intersection with an old plan in which cycle lengths, offsets, and splits were different. Figure 32 shows progression diagrams for all eight coordinated movements along the arterial section. The peak periods are delineated in the figures. While more detail about the quality of progression is revealed in the performance measures described earlier, this view allows a quick qualitative characterization of operation. By visually inspecting the graphs, it is evident that the northbound and southbound coordinated platoons are adequately captured during the AM and PM peaks respectively. The southbound movement at intersection 11 is the exception, because it has random arrivals. Figure 33 shows similar diagrams for operating conditions when the backup controller was used to operate intersection 12. Because this was an older controller, no data was available from intersection 12 during this time. However, the impact on operations at the other intersections is clear from comparing back to Figure 32. For example, we see that southbound coordination during the AM peak was severely disrupted at intersections 13 and 14 (Figure 33, a, b). Northbound coordination at intersection 11 was upset during the PM peak (Figure 33, c). If Figure 33 forms an arterial-level dashboard view, the next level that we drill down to would be the intersection level. The progression diagram for the southbound 14

16 June 1, 29 movement at intersection 13 is shown in Figure 34. This graph shows what took place downstream from the backup controller. The disruption of southbound coordination during the AM (Figure 34, a,b) and PM (Figure 34, e) peaks is evident from this figure. The off-peak (Figure 34, c,d) and evening (Figure 34, f) time periods show interesting patterns caused by a different upstream cycle length. Here, we see patterns of oscillating platoon arrivals in green and red. Drilling down further, Figure 35 shows a magnified view a 45-minute period around 21:. The mainline platoon continuously arrives at an earlier time during each cycle. The arrival of the platoons forms a obvious slope that corresponds to the cycle length disparity. The platoons appear to be arriving at a time in the cycle that is 5 seconds earlier than in the previous cycle; from this, we determine that the backup controller was operating at a cycle length approximately 5 seconds shorter than the rest of the system. The progression diagrams provide a qualitative view of the situation. In this case, removing one controller from the system obviously leads to poorer operation. The performance measures quantify how severe the problem was. Figure 36 shows 24-hour plots of arrival type and Figure 37 shows IO delay in parts (a) and (b) for the northbound movement at intersection 11 and southbound movement at intersection 13 respectively. Arrival type was reduced, indicating a poorer quality of progression at the downstream phases. Coordination at Intersection 13 was especially disrupted, especially during the PM peak. In Figure 34, the coordinated platoon can be seen to arrive during the red interval. This is reflected in Figure 36b by arrival types of 1 3 as opposed to 4 6 during normal operation. Interestingly, the northbound movement at intersection 11 (Figure 36a) did not see as severe of an impact, and actually saw a modest improvement during the AM peak. Because delay for coordinated phases is strongly driven by progression quality, the impacts on estimated delay closely follow the changes in arrival type. As expected, delay was generally higher using the backup controller. Again, we note an exception for the northbound movement at intersection 11 during the AM peak (Figure 37a), which saw improved arrival types. The southbound movement at intersection 13 (Figure 37b) saw much higher delay with the backup controller for most of the day. During the PM peak period, estimated delay increased by approximately 5 times, as arrival types dropped by 3 or 4 categories. 15

17 June 1, 29 Conclusion This paper has discussed a variety of methods of compiling information about arterial performance and provided a number of example views of likely data to be generated at an arterial outfitted with appropriate technology. This information would allow the identification of problem areas with greater precision in time and in terms of specific signal settings than telephone reports and other anecdotal evidence that current traffic engineering practice often utilizes. Information on overall intersection utilization (Figure 8, Figure 9), congestion deficiencies (Figure 1, Figure 11, Figure 13, Figure 15), and quality of coordinated arterial progression (Figure 16, Figure 17, Figure 18, Figure 19) can be inferred from these kinds of reports. Furthermore, at a higher level, this information would allow comparison among facilities across regions managed by an agency in order to assist in allocating resources. Future efforts in this research will seek to define the recommended detector configurations for generation of performance measures, and will expand upon the idea of closing the loop between information reporting and system adjustment by exploring potential suggestions of timing plan changes through examination of performance measure data. 16

18 June 1, 29 References 1. Cambridge Systematics and Battelle Memorial Institute.. An Initial Assessment of Freight Bottlenecks on Highways. FHWA, USDOT, October 25. Retrieved from on March 3, Quiroga, C. and D. Bullock, Travel Time Studies with Global Positioning and Geographic Information Systems: An Integrated Methodology, Transportation Research Part C, Pergamon Pres, Vol. 6C, No. 1/2, pp , US Department of Transportation Research and Innovative Technology Administration, Intellidrive informational website. Retrieved from on March 3, Smaglik, E.J., A. Sharma, D.M. Bullock, J.R. Sturdevant, and G. Duncan, Event- Based Data Collection for Generating Actuated Controller Performance Measures. Transportation Research Record No. 235, Washington, DC: Transportation Research Board, pp , Highway Capacity Manual Day, C. M., E.J. Smaglik, D.M. Bullock, and J.R. Sturdevant, Quantitative Evaluation of Actuated Versus Nonactuated Coordinated Phases, Transportation Research Record No. 28, Washington, DC: Transportation Research Board, pp. 8 21, Smaglik, E.J., D.M. Bullock, T. Urbanik, and D.B. Bryant, Evaluation of Flow- Based Traffic Signal Control Using Advanced Detection Concepts, Transportation Research Record No. 1978, Washington, DC: Transportation Research Board, pp , Sharma, A., D.M. Bullock, and J. Bonneson, Input-Output and Hybrid Techniques for Real-Time Prediction of Delay and Maximum Queue Length at a Signalized Intersection," Transportation Research Record, #235, TRB, National Research Council, Washington, DC, pp , Sharma, A., and D.M. Bullock, Field Evaluation of Alternative Real-Time Methods for Estimating Delay at Signalized Intersections, Proceedings of the 1th International Conference on Applications of Advanced Technologies in Transportation, Athens, Greece, May 27-31, Liu, H.X. and W. Ma, Real-Time Performance Measure System for Arterial Traffic Signals, Transportation Research Record, Paper ID: 8-253,

19 June 1, 29 (a) Freeway system freight bottlenecks with degree of severity. (b) Sorted histogram of bottleneck severity. Figure 1: Example high level network performance measure. ( 18

20 June 1, 29 State District County Int SR 37 & SR 32 Int SR 37 & Pleasant Arterial System Int SR 37 & Town and Country Int SR 37 & Greenfield Ave. Figure 2: The scale of the problem of traffic signal management for regional and local agencies. 19

21 June 1, Figure 3: Travel time on SR 26 in Lafayette, IN. 2

22 June 1, point moving average POG.5.25 Time in Cycle : 2: 4: 6: 8: 1: 12: 14: 16: 18: 2: 22: : Time of Day Figure 4: A progression diagram for the northbound movement at intersection 12. Review Objectives & Constraints Continue Monitoring No Collect Data Generate Performance Measures Evaluate System Performance Intervene? Yes Define Problem Areas and Prioritize Identify Potential Solutions Implement Figure 5: Block diagram outlining a data rich strategy for managing closed-loop arterial signal systems. 21

23 June 1, 29 Table 1: Approximate amount of data to be transferred for a moderately busy actuated coordinated eight phase intersection. (a) Estimate for storing all signal events (6 bytes per event). Event Type Number of Events Number of Bytes Phase Events Phase green 8, 48, Phase minimum green 8, 48, Phase yellow 8, 48, Phase termination code 8, 48, Detector Events Presence Detector On 5, 3, Presence Detector Off 5, 3, Count Detector Pulse 5, 3, Coordination Events Coordination State Change 1, 6, Yield Point 1, 6, Pattern Changes 1 6 Total 184,1 1,14,6 Total for System of 1 Intersections 1,841, 11,46, Agency Total of 2 Systems 36,82, 22,92, Total for 365 day archive (2 signals) 13,439,3, 8,635,8, (b) Estimate for storing first-order performance measures (other performance measures can be derived from these). Object Data Type Objects per day Bytes per Object Total Bytes Cycle Time Timestamp 1, 4 4, Vehicle Count Small Integer 8, 2 16, Green Time Floating Point 8, 2 16, Termination Type Small Integer 8, 1 8, Percent on Green Floating Point 8, 2 16, Estimated Delay Floating Point 8, 2 16, Pedestrian Indicator Small Integer 8, 1 8, Occupancy Floating Point 8, 2 16, Total 57, 1, Total for a System of 1 Intersections 57, 1,, Agency Total of 2 Systems 11,4, 2,, Total for 365 day archive (2 signals) 4,161,, 7,3,, 22

24 June 1, 29 Int SR 37 & SR 32 Int SR 37 & Pleasant Int SR 37 & Town and Country Int SR 37 & Greenfield Ave. Figure 6: Indiana State Road 37 is a signalized arterial system used by this paper to illustrate example system performance measures. 23

25 June 1, Int Int Int Int Int Int Int. 14 Int (a) Total number of vehicles served (b) Total green time provided (s). Figure 7: 24-hour operational statistics. 24

26 June 1, Xc : 2: 4: 6: 8: 1: 12: 14: 16: 18: 2: 22: : Time of Day Figure 8: Degree of Intersection Saturation over 24 hours. The lines show 2-point moving averages. 25

27 June 1, Region of Concern Xc Rank Order (a) All cycles (24 hours) Xc Rank Order (b) Top 1 cycles (24 hours). Figure 9: Sorted Degree of Intersection Saturation. 26

28 June 1, Region of Concern 1.25 Volume-to-Capacity Ratio Rank Order (a) All cycles (24 hours) Volume-to-Capacity Ratio Rank Order (b) Top 1 cycles (24 hours). Figure 1: Sorted Combined Volume-to-Capacity Ratios. 27