Guide to Establishing Monitoring Programs for Travel Time Reliability (2014)

Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

11 This guide describes how to develop and use a travel time reliability monitoring system (TTRMS). It explains why such a system is useful, how it helps agencies do a better job of managing network performance, and what a traffi c management center team must do to establish a TTRMS. Travel time reliability is the absence of variability in the travel times. If a system is reliable, people can get to where they want to go, when they want to be there, all the time. For example, if a freeway is reliable, then its travel times are the same for the same condition, all year long. It is similar to a vehicle that always starts when the key is turned on. Of course, in reality no system or roadway is perfectly reliable, and unexpected events sometimes signifi cantly reduce the reliability of a system. These unexpected events are due to internal and external infl uencing factors like capacity reductions and traffi c control devices (internal) and weather, incidents, special events, work zones, and demand (external). A TTRMS should be created because it provides features and capabilities that most transportation management centers (TMCs) lack. A TTRMS can help operating agencies monitor the performance of their system, understand the impacts of the vari- ous infl uencing factors, provide credible information to system users about what travel time reliability to expect, and decide what actions to take to help improve reliability. Creating a TTRMS involves creating a new module that plugs into an existing TMC platform. The TTRMS relies on the TMC to gather sensor data, manage data processing and storage, and communicate the results of its fi ndings to the outside world. The TTRMS focuses on using the incoming sensor data, along with supplemental 1 INTRODUCTION

12 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY information about the influencing factors, to create a credible picture of how well the system is performing at the present time and in the past. The payoff is better reliability. The TTRMS will ensure that the TMC team knows how reliability will suffer if certain events take place. It will let them understand the impacts of weather, incidents, and special events. It will also help the TMC team man- age the network’s reliability in real time by providing up-to-date information about how the variability in segment and route travel times is either increasing or decreasing in response to actions taken. It is clear that this variability exists. Figure 1.1 shows the various 5-minute travel times observed across 2011 for I-5 in San Diego, California. There are 72,000 data points plotted in the figure. It is easy to see that travel times vary widely during the p.m. peak: • The variation is greater when congestion is higher (i.e., when the demand-to- capacity ratio is high). • The variation is much lower off peak, especially at night and early in the morn- ing, when congestion is low. There were impacts not only from congestion (which is internal to the system) but also from incidents, weather, special events, and unusually high demand. 0 20 40 60 80 100 120 140 160 180 0:00 6:00 12:00 18:00 0:00 Tr av el T im e (m in ) Time of Day (hr:min) No Events Incidents Demand Special Events Weather Figure 1.1. Five-minute average travel times on I-5 in San Diego.

13 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY Figure 1.2 shows the four key information flow steps a TTRMS must execute to fulfill its purpose as a decision support tool. First, the TTRMS needs to effectively measure travel times. This is a complex technical topic due to the variability of traveler behavior and the plethora of different measurement sensors. Correctly measuring travel times along a given route requires a great deal of systems development effort and statistical knowledge. This guide serves as a primer on how to measure travel times effectively using available technologies and statistical techniques. Measuring an individual travel time is the foundational unit of analysis for reliability monitoring. Second, the TTRMS needs to clearly characterize the reliability of a given system. This entails taking a set of measured travel times and assembling them into a statistical model of the behavior of a given segment or route. In this guide, regimes refer to the categories of conditions under which the segment, route, or network is operating at a given point in time (or from one time to another). The statistical paradigm outlined in this guide uses probability density functions (PDFs) to characterize the performance of a given segment or route, usually specific to a particular operating regime (a combina- tion of congestion level and nonrecurring event). This guide gives specific advice on the statistical decisions required to effectively characterize the travel times. Characterizing the reliability of a segment or route is fundamental to making good decisions about how to improve the performance of that segment or route. Third, the TTRMS should identify the sources of unreliability. Once the reliability of a segment or route has been characterized, transportation managers need to under- stand what caused the unreliability (and how to improve it). This guide follows the Low High Computation Engine Figure 1.2. Information flow in a TTRMS.

14 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY causal list used by the Federal Highway Administration to describe why congestion arises and breaks these sources into the seven major influencing factors (two internal and five external) shown in Figure 1.3. It discusses how to pull in data for these influ- encing factors and effectively fuse them with travel time data. Identifying the travel times affected by these sources of congestion is required preparation for understanding system reliability. Finally, the TTRMS should help operators understand the impact of these sources of unreliability on the system. This final step in turning raw data into actionable deci- sions requires both quantitative and qualitative methodologies: operators need clear visualizations of data, as well as quantifications. This dual approach supports both data discovery and final decision making about a given route. Understanding reliabil- ity is the key to good decision making about improving system reliability. Once in place, a TTRMS that accurately and consistently executes these four steps becomes a powerful tool for traffic management. It is a tool that decision makers can use to understand how much of their delay is due to unreliability, and how to mitigate that delay. For example, should a freeway operator deploy more service patrol vehicles (to clear incidents more quickly) or focus efforts on coordinating special-event traffic (to reduce delay from stadium access)? A reliability monitoring system, as outlined in this guide, can tell an operator which of these activities is worth the investment and what the payoff will look like. Building this system will add a new, powerful, practical traffic management tool to the arsenal of system operators. Figure 1.3. The seven major sources (also called factors) of nonrecurrent congestion. Source: Strategic Highway Research Program 2.

15 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY CONTEXT WITHIN STRATEGIC HIGHWAY RESEARCH PROGRAM 2 Reliability is one of the four focus areas of the Strategic Highway Research Program (SHRP) 2, authorized by Congress in 2006, and its purpose is to “reduce congestion and improve travel time reliability through incident management, response, and miti- gation” (1). Four thematic groups have been established under this focus area: • Data, metrics, analysis, and decision support; • Institutional change, human behavior, and resource needs; • Incorporating reliability into planning, programming, and design; and • Fostering innovation to improve travel time reliability. The project under which this guide has been prepared is SHRP 2 Project L02. The project supports the first theme by providing guidance to operating agencies about how they can put better measurement methods into practice and understand the rela- tionship that travel time reliability has to the seven major sources of nonrecurrent congestion (2): • Traffic incidents; • Weather; • Work zones; • Fluctuation in demand; • Special events; • Traffic control devices; and • Inadequate base capacity. A key part of the delivery of travel time reliability to the traveling public is the presence of a TTRMS that can assist engineers, planners, managers, and policy makers in making sound traffic management decisions. STRUCTURE OF THE GUIDE The Guide follows the block diagram presented in Figure 1.4 to assist in describing the TTRMS. Each module is shown as a box, and the inputs and outputs are shown as circles. “Other Systems” refers to systems that monitor the influencing factors. Five chapters define and describe the TTRMS: • Chapter 1, Introduction, presents an overview of travel time reliability. • Chapter 2, Data Collection and Management, discusses the types and application of various types of sensors, the management of data from those sensors, and the integration of data from other systems that provide input on sources of unreliability (e.g., weather, incidents). This represents the left side of the block diagram in Fig- ure 1.4 and includes traffic sensors, other systems, and the data manager.

16 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY Figure 1.4. Reliability monitoring system overview, with boxes for modules and circles for inputs and outputs. • Chapter 3, Computational Methods, discusses how PDFs are derived from the various data sources. This represents the center part of Figure 1.4 and includes the process of generating travel time PDFs (TT-PDFs) that can be used to derive a variety of reports to users. • Chapter 4, Applications, presents a discussion of potential use cases for a travel time monitoring system and illustrates its use with a series of real-world case studies. • Chapter 5, Summary, illustrates the entire analytical process of a TTRMS. The Guide is supplemented by four appendices that provide additional detail to support the development and application of travel time monitoring systems: • Appendix A, Monitoring System Architecture, provides supporting detail for Chapter 2 and presents examples of detailed data structures for the organization of various data sources. • Appendix B, Methodological Details, provides supporting detail for Chapter 3 by presenting detailed discussions of the analytical methods that can be used to calcu- late travel time reliability measures from a variety of input sources. • Appendix C, Case Studies, provides supporting detail for Chapter 4 and offers detailed case studies that exercise various aspects of the Guide, including system architecture, analysis of recurrent and nonrecurrent sources of congestion, and the application of a variety of use cases. • Appendix D, Use Case Analyses, illustrates the application of a variety of use cases for a TTRMS to provide further supporting detail for Chapter 4.

17 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY HOW TO USE THE GUIDE Agency staff can read this guide to understand what a TTRMS can do and what fea- tures make the most sense for their own situation. For example, from which sources can an agency collect data? Which steps can an agency employ in processing data? What performance measures can an agency realistically calculate? How can informa- tion be presented to various end users? This guide can be used as a starting point for agencies to review current practices, recent reliability research, and developing tech- nologies with the aim of designing a system that meets each agency’s particular needs. The use cases and case studies presented in Chapter 4 helped guide the prepara- tion of the module descriptions. Use cases are a formal systems engineering construct that transforms user needs into short, descriptive narratives that describe a system’s behavior. Use cases capture a system’s behavioral requirements by detailing scenario- driven threads through the functional requirements. The collective use cases define the monitoring system by capturing its functionalities and applications for various users. Each use case defines the purpose for the analysis, the steps the user must take to obtain the desired data, the results of the steps, the spatial aggregation used, and the travel time reliability measures that can be reported. The measures listed in these use cases are only examples and represent possible ways that reliability information can be conveyed. Travel time monitoring increasingly incorporates a blend of infrastructure from the public and private sectors. As a result, some elements of travel time monitoring systems in current use have proprietary components. This guide is not intended to pro- vide a bid-ready specification of all aspects of a travel time monitoring system, some aspects of which may be proprietary. Nevertheless, the reader will be able to better understand the functional specification of such a system as defined by its required inputs and outputs. WHAT IS TRAVEL TIME RELIABILITY? While no consensus exists on the definition of travel time reliability, Elefteriadou and Ciu (3) provide a starting basis in their study of travel time reliability definitions. As they comment, Ebeling (4) provides a useful basic idea, defining reliability as “the probability that a component or system will perform a required function for a given period of time when used under stated operating conditions. It is the probability of a non-failure over time.” Reliability is thus slightly different from consistency, which has to do with the absence of variability. In a highway network context, a system is reliable (formally speaking) if each traveler or shipper experiences actual time of arrival (ATA) that match desired time of arrival (DTA) within some window, as shown in Figure 1.5. In some cases ATA and DTA are extremely important, depending on trip constraints and penalty at the desti- nation. For example, a trip to catch a plane is more constrained than a trip to the store. If the ATA lies outside the DTA window, especially if the ATA occurs after the DTA, a reliable trip was not completed. That is, the transportation system is reliable, technically speaking, if the ATAs all lie within their DTA windows. Otherwise, the

18 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY system has failed. As Elefteriadou and Ciu (3) point out, such a definition of reliability becomes well-defined. In a general sense, the reliability of the system can be measured using utility theory, as described, for example, by Hansson (5). Utility (of the trip) is maximized if the ATA is inside the DTA window; conversely, disutility is greater if the ATA lies outside the DTA window. The aggregate disutility for all trips among all users is the “societal cost” of the system’s unreliability. The function that evaluates the disutility may be symmetric or asymmetric depending on the situation, as shown in Figure 1.6. Truckers incur significant penalties if they are either late or early in delivering shipments to receivers. Individual travelers can be late for appointments or miss the opportunity to insert additional tasks like stopping for coffee or sleeping later if they are early. If it were possible to observe all the trips on a given transportation system, and if all DTA windows for those trips were known, then the reliability of a given transpor- tation system could actually be assessed. One could compute the percentage of ATAs that fall within their DTA windows. This would be a useful metric both for the entities making the trips and the organizations providing the service (e.g., the TMC or transit system operator). The aggregate disutility could also be computed by summing the disutility values for each trip. Obviously, this world does not exist. What can be observed today, at least in part, are travel times on segments and routes in the network. For example, some TMCs can monitor probes (vehicles equipped with tags in areas that have toll roads) and others can generate speed distributions at specific point locations in the network where sensors (speed traps) are installed. As a result, the TMC can establish desired travel times or, better yet, desired travel rates (DTRs) in seconds per mile so that the length of the facilities does not interfere with Figure 1.5. Concepts of desired and actual times of arrival.

19 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY performance levels to be achieved on the segments and routes, consistent with Ebeling (4). These DTRs can be dependent on the regime under which the system is operating (combination of the influencing factors), and they can be adjusted over time as net- work conditions change (e.g., as demand grows or improvements are made). If single-value measures seem easiest to master or communicate, the median travel rate is a good choice. If more complex and powerful tools are desired, travel rate PDFs (TR-PDFs) are a good choice. Research conducted for the development of this guide found that in general, travel times do not follow a normal (Gaussian or bell-shaped) statistical distribution. As a result, the median value (half above and half below) is generally a better measure of central tendency than the more familiar average (mean) value. Similarly, semivariance was found to be a better measure of variation than the standard deviation. A segment or route is performing reliably if its actual travel rate (ATR) lies within the acceptable DTR window given the regime under which the segment or route is operating. The TMC team can monitor the number of segments or routes whose ATR lies within the DTR window; they can see how that number varies based on the net- work, segment, or route operating conditions (e.g., an incident during high conges- tion); and actions can be identified to increase the number of segments or routes whose ATRs are within their DTR windows. This paradigm can also be extended to the system users. Trips can be considered successful if their ATR falls within an allowable DTR window based on the condi- tions under which the trip was made. Reliability can be measured by the percentage of trips whose ATRs fall within the allowable DTR window. By extension, the aggregate disutility experienced by the travelers or shippers can be assessed, in principle, using Figure 1.6. Disutility function to characterize desired and actual times of arrival.

20 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY disutility functions that compare the ATRs one at a time with their corresponding DTRs and then sum the results. Service providers want to see if different ways to operate the system would be likely to produce better alignment between the ATRs and DTRs (or if capacity invest- ments are needed). Naturally, this decision making is aimed at variability reduction and shifts in the median values, either lower or higher, so that the requisite confidence interval objectives are met given the desired travel time windows. The decision making by the TMC team is similar to the mean–variance trade-off analyses so prevalent in financial planning; see, for example, Maginn et al. (6). In this instance, the trade-off is between minimizing the median travel times by, for example, building new network links or adding capacity to reduce the median travel rates, ver- sus taking actions such as improving incident response or managing the impacts of weather better so that the variation in the travel rates is reduced (i.e., getting more ATRs within their DTR windows). SUPPLY-SIDE AND DEMAND-SIDE PERSPECTIVES The concepts of supply side and demand side have some bearing on travel time reli- ability. The supply side focuses on the TMC team and its interest in managing the seg- ment-, route-, and network-level reliability so that performance goals are met. These goals are most likely predicated on simple metrics like the median travel rate rather than the distribution of individual vehicle travel times. In addition, consistency across time (in a longitudinal sense) is the primary concern: how does the team make sure that every a.m. peak has an adequate reliability performance? The demand side focuses on individual travelers and the travel times they expe- rience in making trips. Travelers’ decisions regarding modes, routes, and departure times are associated with their own sociodemographics and the context in which they travel. Observations of individual vehicle (trip) travel times are important, and the travel time density functions of interest pertain to travel times experienced by indi- vidual travelers at specific points in time. That means the reported percentiles of those travel times (e.g., the 15th, 50th, and 85th percentiles) pertain to the travel times for those travelers. When considering reliability it is important to understand the context of the analysis. Two analysts could think they are talking about the same reliability ideas. One, however, may be focused on the reliability of median travel times for the same 5- minute time periods across a year (the duty cycle for the network), while the other is focused on the travel times experienced by individual vehicles in those same 5-minute periods across the year, or for a single 5-minute period (or an hour) for a given net- work on the same day (the latter is more of a cross-sectional data analysis). Neither analyst has the right or wrong perspective; both are valuable. However, before they go too far in their discussion, they need to make sure they are both thinking about the problem in the same way or clarify that they understand how their perspectives are different and relate to one another.

21 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY TRIP-MAKING CONCEPTS Several trip-making concepts are important in understanding the material presented in this guide. The first and most important concept is that trips made by persons and packages are the fundamental events on which travel time reliability is focused. Per- sonal travel time reliability has to do with individual person trips, and freight travel time reliability has to do with individual package shipments. The second concept is that travel times and trip times are different. A trip refers to an origin, a destination, and implicitly, a trip purpose. Trips can be chained trips involving intermediate stops, and thus trip times can be different from travel times. Trip times are the end-to-end times involved—from origin to destination—and allow for intermediate stops and side-trips. Travel times have such extra times removed and refer to the amount of time spent traveling. However, this distinction also pertains to observations of times for segments and routes. Trip time is the elapsed time between when one sensor identifies a vehicle and a second sensor at some other location identifies the same vehicle at a later time. Trip times can be travel times, but not necessarily. The trip time could include stops or detours, since the specific route of the vehicle between the two sensors might not be known. Travel time refers to the actual driving time between the sensors. Trip times need to be filtered in order to obtain accurate travel times. The third concept is travel rate. This is travel time per unit distance, such as min- utes per mile. Travel times help travelers understand how long it will take them to accomplish their trips and help system managers determine what travel times to dis- play on variable message signs. In contrast, travel rates help system managers compare the performance of one segment with another so that strategies can be developed for making capital investments that help to improve the performance of the segments and the network as a whole. The fourth concept—an acceptance of the current technological frontier—is that individual person and package trips and travel times are difficult to observe. At present the best options available are automatic vehicle identification– and automatic vehicle location–based sensors. When those technologies become more prevalent, it will be easier to collect data on individual trips. The final concept is the acceptability of using aggregate measures as surrogates for the more detailed picture of individual trips. The aggregates are derived from the more detailed information (e.g., as in the average speeds provided by single-loop and double-loop detectors). Also, single-occupant vehicles are a significant portion of the traffic stream, especially during congested conditions, so insights about individual per- son trips can be seen in the vehicle data. The main point is that travel time reliability and the ideas about it presented in this guide are intended to address questions about individual person and package trips, even though surrogate data are used. This means the methodology is designed to provide insights about travel time reliability that might be experienced by persons and packages by extension of the insights derived from analyses of more aggregate measures.

22 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY PROBABILITY DENSITY FUNCTIONS Much of the material presented in this guide emphasizes the creation and analysis of PDFs to describe the distribution of travel times and travel rates and to compare the performance of one facility with another or one operating condition with another. These distributions are often presented three ways. The first way is as a histogram, in which bar heights are used to represent the relative frequency with which specific conditions pertain. Figure 1.7 shows a histogram of travel times for I-8 in San Diego during the a.m. peak for various operating conditions: normal, when the system was affected by an incident, and when the system was affected by weather. No special events were observed in the selected data set. Figure 1.7. Travel time histogram for various event conditions. )

23 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY The second way in which these distributions are presented is via a PDF. Figure 1.8 shows the corresponding PDFs for the three conditions. The PDF is a density function that is based on the values in the histograms. In the PDF, as in the histograms, one can clearly see that some travel times are more common than others and that the distribu- tion of the travel times is different for the various operating conditions. Note that these histograms of San Diego data do not follow the familiar bell-shaped curve. The data have a significant skew even after they have been categorized. The third way in which these distributions are presented is in a cumulative density function (CDF). The CDF is based on the PDF in that a value shown in the CDF at any point in the graph is the integral of the PDF up to that point (i.e., the area enclosed within the PDF above the horizontal axis). A property of the PDF is that its area sums to 1.0, which means the CDF ultimately rises to a maximum of 1.0. Figure 1.9 shows the CDFs for the various regimes associated with the performance of I-5 in San Diego (the same facility illustrated in Figure 1.1). As with the PDF, one can clearly Figure 1.8. PDFs for various event conditions.

24 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY Figure 1.9. CDFs under various regimes (operating conditions). Uncong = uncongested. see differences in the distribution of the travel rates (as compared with travel times) and that the distribution of rates varies among regimes. It is through the use of these tools—the histogram, PDF, and CDF—that the case studies and use cases reach conclu- sions about the influence of various factors on the travel times and travel rates. The distributions can be, and often are, multimodal (in a statistical sense, meaning there is more than one point at which the PDF reaches a maximum). This can occur because the data may reflect more than one operating condition or regime across the span of time being studied or because users may have experienced different treatments during the analysis time frame. An illustration of the former arises when the average travel times for a given 5-minute time slice are studied across a year. It is likely that dif- ferent operating conditions existed on different days (because of weather and incidents and different demand levels), so multimodality should be expected. An illustration of the latter situation occurs when travelers do not all experience the same control treat- ment. For arterial networks, this could arise when some users are able to progress between traffic signals without stopping, but others are not. For freeway networks, it can be because some vehicles are delayed by ramp metering controls, but others are not; or when some vehicles experience delays from paying tolls (e.g., cash), but others do not (e.g., toll tags).

25 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY Since the word mode is used in other ways in transportation, in this guide the word regime is used instead of mode to describe these various operating conditions (or subconditions). Moreover, common traffic engineering names (e.g., congested, uncongested, transition, incident, and weather) are used to describe these regimes. The regimes help enhance the quality of the PDFs. They keep the PDFs from being noisy, and they help maximize the incremental value derived from data acquired every day. The last concept—and an important one—is that all the reliability metrics of inter- est can be derived from these PDFs. The PDFs completely describe the travel times or travel rates (travel times per unit distance). Hence, the typical metrics of interest for characterizing reliability—planning index, buffer index, average, median, 95th per- centile, or others—can be computed on the basis of the PDFs. As a result, these PDFs, supplemented by ancillary data about the environment (e.g., weather, incidents) that exists (or will exist) in the time frame of the analysis, represent sufficient information to answer the questions about measuring reliability. REFERENCES 1. Reliability Focus Area Overview: Providing Reliable Travel Times on the Highway System. Strategic Highway Research Program 2, Transportation Research Board of the National Academies, Washington, D.C. June 17, 2009. http://onlinepubs.trb. org/onlinepubs/shrp2/RRPJune2009.pdf. Accessed Sept. 2, 2012. 2. Cambridge Systematics, Inc., and Texas Transportation Institute. Traffic Conges- tion and Reliability: Trends and Advanced Strategies for Congestion Mitigation. Federal Highway Administration, Washington, D.C., Sept. 2005. 3. Elefteriadou, L., and X. Ciu. Review of Definitions of Travel Time Reliability. Presented at 86th Annual Meeting of the Transportation Research Board, Washington, D.C., 2007. 4. Ebeling, C. E. Introduction to Reliability and Maintainability Engineering. McGraw-Hill, 1997. 5. Hansson, S. O. Decision Theory: A Brief Introduction. Royal Institute of Tech- nology, Stockholm, Sweden, 2005. 6. Maginn, J. L., D. L. Tuttle, D. W. McLeavey, and J. E. Pinto. Managing Invest- ment Portfolios: A Dynamic Process, 3rd ed. CFA Institute Investment Series, John Wiley & Sons, 2007.