Scoping and Conducting Data-Driven 21st Century Transportation System Analyses
Conducting a time-dynamic operational simulation analysis of a transportation system was a novel endeavor 35 years ago. With personal computers, continuous loop detector data, and new software tools, codifying basic driver behaviors enabled analysts to understand and optimize the transportation system. A new capability emerged—we could examine a transportation system as it evolved over time over the course of a day. Analysts could consider animations of vehicles approaching and departing intersections, choosing lanes, and accelerating and decelerating to maintain safe distances across a virtual roadway. Structured analyses could examine the onset, nature, and duration of congestion within a virtual system in a way that resonates with the actual experience of the transportation system user.
New notions of optimal performance influenced by travel time and travel-time reliability became central concepts of system management. The environment for the 20th Century pioneers was exciting and intellectually challenging, but these emerging analytic capabilities had significant limitations for broad practical application to support transportation system management. Software tools were in primitive states of maturity, often derived from academic research efforts. Traveler behavior models used simple relationships based on what limited data were available, or filled in with best-available assumptions where empirical experiments had not been conducted. Field data to calibrate models of roadway systems were often scarce, incomplete, and riddled with errors. Computer platforms were unable to support the representation of large, complex and multi-modal networks—or required impractical amounts of time for these models to run to completion.
Since the early days of exploration and invention, there has been an evolution in the environment for 21st Century transportation systems analysts. This transformation has been driven by, among other factors, improvements in commercially supported software, low-cost access to advanced computational platforms, and a data-rich environment. Conducting some less complex experiments has evolved into standardized practices that inform some routine decisions (e.g., intersection warranting and interchange justification analyses). Forward-thinking analysts have attempted to push conventional boundaries of network scope and analytic purpose (e.g., multi-modal analyses and corridor and sub-regional analyses). While a history of partial successes and near-failures can be recounted, collectively, the field eventually cast aside a number of these limitations. Now, the scope of operational analyses is much larger than it was in the past—and the demands on analysts to address more difficult questions with even more complex modeling approaches continue to grow.
While these technological changes have empowered the 21st Century analyst, the gains made by the individual analyst may have outstripped the ability of the organizations managing transportation systems to capitalize on new analytic techniques. The transportation analyst began work in isolation as a researcher/super-user; to some extent, this notion of isolated "nerds in the basement" conducting analyses still persists—with analysts ensconced in specific consulting firms, analytic departments, and graduate school laboratories. A sociologist might easily distinguish a subculture of modeling and simulation experts gathered around a specific tool and associated networks, largely insulated from the mission of improving surface transportation system performance. The analyst is rarely involved in diagnosing transportation system problems or using data to support analytic project scoping. Isolated champions with the vision and institutional connections to bring analytics to bear in decision-making have made important contributions, but these connections tend to last only as long as the champion remains engaged. With rare exceptions, there is a lack of advanced institutional models to systematically and consistently leverage the power of transportation analytics embedded within the broader transportation system management mission.
Funding transportation analysis can provide valuable insight into the potential benefits and costs of transportation investments. The general value of conducting analysis is the extent to which it assists stakeholders to:
Objectives of This Document
This document provides guidance on how 21st Century transportation analytic resources—data, tools, and computational platforms—can be systematically embedded within the transportation system management process, resulting in:
Transportation system management organizations that successfully achieving these objectives will realize a number of important positive outcomes related to greater insight, better analyses, and reduced costs and risks:
Definition of Operational Analysis
This section defines the scope of an operational transportation systems analysis. Operational analyses are inherently time-variant, with explicit treatment of time within an analytical tool rendered in time steps bounded above by an hour and below by fractions of a second.
An operational analysis, therefore, includes all traffic micro- and meso-simulations, some time-dynamic macro-simulations, dynamic transit operations models, activity models, time-dynamic applications of statistical tools, prediction constructs, communications models, and integrated modeling frameworks combining these classes of tools.
Why are time-dynamic analytics critical? Because a busy modern transportation system is inherently a dynamic entity. It never exists in pure equilibrium, and is in a state of perpetual change at multiple temporal wavelengths: minute-to-minute, hour-to-hour, peak-to-nonpeak, day-to-day, seasonally, and year-to-year. A time-dynamic view of the network informs a fundamental element of transportations systems management, namely, that system management is essentially a task of managing change. Without time-dynamics, the view is critically skewed, and all decisions made regarding transportation system management will suffer from what is essentially a self-limited, stunted understanding of the true nature of the system.
The time-dynamic view most closely represents the way the traveler (or system user) experiences the transportation system. For the system user, each day represents a risk in missed appointments and wasted time. For many system users (both travelers and those who move goods from place to place), predictability of travel and the reliability of travel over repeated interactions with the system is paramount. The time-dynamic system view captures both the within-the-day experience of travelers seeking to use the system without excessive delay and the longer-term experience of system users seeking to manage uncertainty and variability in system performance.
Influence of Operational Analyses
Operational analyses can inform decision-making on multiple time scales: real-time, operational planning, planning for operations, and long-range planning. The incremental time-step within the time-dynamic representation of the network may be very small compared to the scale of decision-making. Understanding how congestion arises, propagates, and subsides within a system can inform decisions made in the next ten minutes, operational changes to be deployed over the next ten hours, investments made to improve operations over the next ten months, or major improvements to be phased in over 10 years or more. Later in this introduction, this guide will identify ways that operational analyses inform decisions at various time scales; we use examples throughout the document to show the breadth of transportation system management decision-making, potentially enhanced with time-dynamic analyses.
There are three critical aspects of the resources available to the 21st Century transportation systems analyst: new data sources and data volume, more powerful computational platforms, and increasingly complex and capable visualization and analytical tools.
The 20th Century analyses were constrained by the data, computational platforms, and tools available at the time. The abstraction of the system required to perform predictive analytics—particularly with respect to time-dynamics—were relatively severe. Lack of data forced analysts to focus on average data from disparate sources and create time-variant "normal" condition days that by definition excluded outlier days with incidents, weather or unusually high (or low) travel demand. For the largest geographic scope (e.g., four-step planning modeling), time-dynamics were removed all together.
With 21st Century capabilities, the analyst can cast aside severe abstractions to realistically and accurately represent the time-variant system in a new and powerful way. Continuous data collection captures outlier conditions as readily as any other operational condition. Computational platform improvements allow more runs in the same amount of time, rendered in higher detail—detail that can be analyzed and visualized in increasingly varied ways to serve multiple needs and provide insights. This richer representation of time-dynamics in the system allows the analyst to understand the transportation system the way a user experiences the system—over time, in repeated trips of varying nature, with good days and bad days—and can characterize performance across all conditions (both good and bad).
Compared to the world 35 years ago, current analyses are conducted in a data-rich environment. Sources of data have expanded from infrastructure-based detection systems to include probe data (vehicles or mobile devices). More data are continuous—available at periodic intervals around the clock—rather than associated with a one-off data collection effort. More data sources reveal the system condition from the user-perspective, expanding the potential views from the traditional collection of facilities (infrastructure-based) to a collection of trip-making activities (user-based). Looking forward, there is no indication that these broad trends will decline, given the increasing number of data sources, more rapid data delivery, and a broader array of types of data. The advent of connected vehicle technologies, the Internet of Things (IoT), and an increasingly engaged and connected traveler will create new opportunities to understand system dynamics from multiple vantage points. Most importantly, these data sources allow analysts to characterize system performance in new and useful ways. Rather than simply examining average delay, they can characterize travel reliability, on-time performance, trip predictability, and other measures that capture the full range of options and experiences users encounter.
In 1965, Gordon Moore, co-founder of Intel, authored Moore"s Law, a prediction that the computational power of analytical platforms will double every two years. (Moore"s Law and Intel Innovation.) Fifty years later, this observation still holds true. The desktop or cloud-based computational tools at an analysts" disposal are more powerful and more ubiquitous. (Moore"s Law Keeps Going, Defying Expectations.) That said, legacy transportation software packages are not inherently high-performance computational tools, and tend to be single-processer–based; the computational power available to the analyst may be limited in many cases since scalable solutions are not directly realized in most analytical tools. In this case, the improvement in power will be largely realized in data storage and visualization, which is inherently more scalable than discrete time-step simulations, for example.
Increasingly complex tools will be increasingly capable. However, the time required to master and understand these complex tools has also increased. As the complexity grows, it becomes harder to manage and estimate all the detailed parameters needed to drive the tools effectively. Analysts must master individual tools at the same time that they understand the strengths and weaknesses of various tool types. In some cases, a simpler tool may actually provide greater insight than attempting to model at the smallest time step with the most parameters in play.
The previous two sections describe how to define time-dynamics and why that is important. The other key concept to be clear about is the system—the thing that is to be managed, modeled, and—hopefully—optimized.
Why is a System-View Important?
A typical pitfall for analytic projects is that the underlying nature of the problem is not well known until the project is over. Often, only near the end of the project does the analyst focus on what specific system is being managed, modeled, and optimized. When it is unclear among multiple stakeholders what system is being modeled, the analyst must often "fill in the blanks" and define the system in modeling. An analytic project may be presented as a pure freeway merge/weave analysis, but it turns out that signal timing on nearby arterials produces dense platooning on the on-ramps that essentially create the merge/weave issue. The "system" here is the combined freeway-arterial roadway system, not just the freeway. An analysis of the freeway alone misses the essential point of the dynamics of the combined system. This example is quite tactical, but there are similar examples regarding individual elements of integrated multi-modal corridors, sub-regional analyses, and multi-state freeway corridor analyses.
What is the System?
The system is the collection of facilities, fleets, infrastructure, and trip-making users for which the system manager is responsible plus all interacting systems that influence the performance of the system for which the manager is responsible. When defining the system, it can be useful to examine a number of boundaries:
Who Owns the System—What is the Span of Control?
After defining the system, it may become clear who really needs to be included in order to adequately address systematic issues and improve system performance. Once the full "system" is defined (in Module 4), we may end up with a new system concept with no single owner; this is highly typical for complex surface transportation systems that must interact. The presence of multiple overlapping spheres of control is a critical factor in nearly all system analyses. What gets optimized in each system relates to overall performance for everyone, but what is good for the goose may not always be thought of as good for her neighbor, the gander. Even if there is no clear system czar, there still can be an understanding of how systems interact. Collectively, managing the broader system is always preferable to uncoordinated local optimization.
Analysts can play a critical role in assembling a community of organizations and decision-makers that manage the shared system. They can help frame the system concept—using the system definition to bring together the community—and show interactions within and across boundaries. They can also use data to visually describe the system, which can be a powerful tool in energizing stakeholders by tangibly illustrating the concept of a collectively managed system. Analysts often use time-dynamic tools to put the conceptual system into motion, showing when and where the system performs poorly, and engaging stakeholders to work together to provide more effective collective system management. A notional system is much more powerful if it can be clearly defined, observed (using data), and studied (using both data and tools).
System of Systems Effects
One example where a system was positively influenced by analytics regards a mixed freeway/ arterial corridor where the main arterial route had multiple jurisdictions, each in control of a particular subset of all the intersections in the corridor. Everyone agreed that congestion in the corridor, particularly on the arterial route, was a serious issue. Arterial travel had become so congested and so unpredictable that travelers complained and small municipalities along the route were worried that shoppers were avoiding doing business in their area because of the hassle of dealing with the congestion. The state department of transportation organized a stakeholder group and initiated a micro-simulation analysis. Only after the model was constructed and some initial runs conducted did it become clear that many signals in the system were optimized for access to/from shopping along the route, rather than for travel along the route, particularly from large parking facilities. Since the policies of these local sub-systems were the largest influencers of arterial performance, adapting other "major" intersection signal control would have limited influence. A workable solution was found by harmonizing turning movement release at major intersections to coincide with demand at/around the major parking facilities. This insight came somewhat late in the effort, so the original analytical model had to be extensively reworked and augmented to reflect the nature of the interacting systems. The system turned out to be highly dependent on the large parking facilities, and effective management of the system relied in large part on incorporating time-dynamic effects of this previously ignored parking sub-system.
This guidance supports transportation professionals—including data managers and transportation analysts—at different levels of technical expertise and in a wide variety of uses across short-, medium- and long-term decision horizons. To effectively identify, scope, and conduct data analytics projects, a project scoping cycle provides a set of standard data-driven processes to reveal insights effectively (and cost-effectively). The defined cycle of steps in the 21st Century Analytic Project Scoping Process underscores the limitations of the past and offers a new way forward as a part of a Continuous Improvement Process (CIP). The process provides a common data-driven analytic framework adapted from modern systems engineering that is designed to enable the demands of modern reliability-focused analytics and alternatives analysis.
Data managers often fall into the trap of capturing, assembling, and managing data without a comprehensive understanding of the underlying analytical uses for these data or understanding that data itself can yield insights when properly integrated. Therefore, the scoping process begins by characterizing key system performance measures that reveal underlying transportation system dynamics, shifting day to day and year to year, subject to many external influences. Data needs are derived both to calculate these measures and to provide the context for the variation (e.g., variation in demand, incident patterns, and weather impacts). Data are then readied for analysis through a focused and resource-sensitive quality control process. These data also drive a series of characterizing analytics that inform and shape the project-scoping activity.
Project scoping includes problem statement and hypothesis development, spatial and temporal dimensions, and tool selection—choosing from the full gamut of potential analytical methods from statistical analysis to various alternatives among simulation techniques. Analytics includes creating a detailed experimental plan that accurately reflects a control (baseline) and experimental (alternative) structure; this plan is evaluated over a level playing field derived from a representative set of varied operational conditions.
Documentation of the data and analysis results is the key component to complete the analytics project cycle and to prepare for the next cycle. The next cycle stage is to circle back to identify what can be statistically derived from the analytics and how these results inform a new set of data needs and further analytic work. Figure 2 (on the following page) illustrates the 21st Century Analytic Project Scoping Process detailed in this guidance, a four-module CIP:
(Source: Federal Highway Administration.)
Module 1 discusses system performance measures and early diagnostic activities to help analysts develop preliminary analytic problem statements and prioritize the identified problem statements with a risk-reward project screening approach. The final product of this module—one or more individual Analytical Problem Statements associated with a high-priority concept—serves as the foundational connecting document for Module 2: Scoping.
Project scoping includes a more detailed project definition, the identification of project-specific performance measures, a refinement of mitigation strategies and data needs, tool selection, and cost and schedule estimation. A Scoping Tool to help the analyst complete this final step is included as a part of this guide. (The scoping tool is available at the U.S. Department of Transportation Open Source Application Development Portal (OSADP) Web site.) Module 2 ends with the completion of the Project Scoping Summary, which provides enough information to initiate data preparation and analysis work (Module 3).
A data analyst in Module 3 ensures the consistency and quality of the data available and outlines the data collection plan to fill any gaps between data needs and availability. Depending on the nature and scope of the project, the data analyst may analyze the available data to identify and summarize a representative set of operational conditions, critical to the creation of a strong analytical plan in Module 4.
In Module 4, an analytical project is executed. The analyst creates a detailed experimental design with experimental and control cases, then calibrates and validates models under various operational conditions and conducts a sensitivity analysis. Documentation of analysis results is the last step—to document project findings that inform decision-making and to capture lessons learned to improve the agencies implementation of the 21st Century analytic project scoping process itself for the next cycle of analytic work.
In the center of Figure 2, the 21st Century analytic process includes maintenance of analytical capital built up from multiple projects. These can be simply thought of as improved analytical tools, system models created as inputs to these tools, and the data collected to assess system performance or characterize operational conditions. However, a third, often-overlooked element is the insight gained along the complete life cycle of the analytical project with respect to lessons learned in all four steps: diagnostics, scoping, data preparation, and analytics. If the individuals performing these steps do not record key insights and move on to new roles in other organizations, these insights are nearly always lost and a new generation of analysts are likely to repeat mistakes made in future analytic work.
Data-driven analytics enable improved decision-making at different time scales, short-term (transportation system monitoring and real-time operational management), medium-term (Transportation System Management and Investment Planning) and long-term (transportation system planning and long-term trends). Each of these time scales place different demands on the data manager and have distinct data needs.
Transportation System Monitoring and Real-Time Operational Management
The cycle of transportation system monitoring and real-time transportation operational management projects can be conducted in real time—every minute, hourly, or daily, depending on the scope of the project. Given the short feedback timeframe, most settings need to be automatic or semi-automatic. Examples are ramp metering and traffic signal control systems. The system can collect and store real-time data and built-in algorithms to conduct data analysis and can store the analysis results. A routine self-diagnostic function can be made automatically for real-time system adjustments or manually on a daily or weekly basis for short-term system adjustments.
Transportation System Management and Investment Planning
From the transportation system management aspect, it takes days, weeks, or months to repeat the project cycle. A work zone alternative analysis project—a type of transportation system management project very similar to the transportation system monitoring project—takes more time to complete the cycle. The system collects daily time-dependent traffic data and compares day-to-day traffic patterns, weather, and incident impacts with different alternatives. The system can identify the best alternatives based on the previous analysis results and current operational conditions. The data and the analysis results are stored properly for the next round of work zone activities to both save the cost and gain lessons learned. It could be documented as a standard procedure when conducting a work zone alternative analysis.
Transportation System Planning and Long-Term Trends
This long-term category includes one-to-three year transportation system planning (e.g., a freeway lane expansion) and over-ten-year long-term trends (e.g., Metropolitan Planning Organization"s [MPO"s] Metropolitan Transportation Plan [MTP]). The Federal Highway Administration (FHWA) published a guidebook of a data-driven, strategic approach for Departments of Transportation (DOTs), MPOs and other planning organizations to make investment and policy decisions to attain desired performance outcomes for the multimodal transportation system. (Federal Highway Administration, MODEL LONG-RANGE TRANSPORTATION PLANS: A Guide for Incorporating Performance-Based Planning, August 2014.) The data-driven/performance-based transportation plan includes a discussion of conditions and performance of the transportation system over several years. As this diagnostics occurs through multiple cycles, the plan serves as a baseline for developing and refining plan goals, objectives, and targets. Information from the performance report can support refinement of targets associated with the timeframe of the transportation plan, as well as near- or mid-term targets.
Relations between Different Time Scales
The MTP is a blueprint that contains a 20-30 year planning horizon and is updated every 3-5 years to identify current and future transportation corridors, and forecast transportation demand, needs and growth patterns. Based on MTP, a 3- to 5-year State Transportation Improvement Program (STIP) or Transportation Improvement Program (TIP) is developed to identify the highway and transit improvement plan and identify short-term priorities with funding sources. The performance targets set in the long-term plan are evaluated during both short-term and mid-term cycles. The analysis results from short-term and mid-term cycles also feed into future long-term plans as part of the system diagnostic process. Figure 3 illustrates interactions among these three time scales. The solid lines represent the project identification and development from transportation analysts. Dashed lines represent insights gained and quantitative feedback from relatively shorter-term analyses that influence relatively longer-term project identification and development.
(Source: Federal Highway Administration.)
A freeway segment expansion is identified with projected performance targets in the TIP based on the demand growth and need prediction in MTP. This decision then goes to the mid-term transportation system management plan to conduct the alternative analysis for the freeway construction period. During the construction period, the short-term system monitoring and real-time management takes place to adjust the system in real time or a short time period. Daily, weekly, or monthly system performance and adjustment reports serve as a basis to exam and adjust the system management plan for the next few months. The system management plan documents lessons learned and key findings serve the future TIP and MTP in the next scoping cycle.
The following is a list of resources related to the planning, development, and implementation of an Intelligent Transportation Systems (ITS) project in the 21st Century. Each resource includes the date if available and brief summary of its content:
A complex transportation system—such as an interstate freeway corridor or a transit system—involves multiple jurisdictions (e.g., state and local agencies or a state highway administration and a transit administration) and multiple stakeholders (e.g., private companies, public agencies, and academic/ research institutes). One of the major challenges in managing such a system is the amount of effort spent coordinating stakeholders. For example, from a state point of view, optimizing a corridor including the surrounding neighborhoods may be a centerpiece of concepts to improve corridor performance. However, for local agencies, such as a county or city, local signal timing optimization is related only to localized (sub-system) optimization. System-wide optimization may be inimical to local optimization and vice versa.
Note that sub-system optimization can be technically challenging but potentially counter-productive. For example, ramp metering controls may seek to increase throughput on the freeway mainlines—even if this implies extra delay at nearby arterial intersections. Similarly, the arterial signal optimization may not take into account impacts on freeway throughput.
Managing the system does not mean running roughshod over the goals of some stakeholders to benefit others. Collective system management includes finding common ground regarding goals for the system and showing why ad hoc local sub-system optimization is worse for everyone than coordinated, optimized system management. Armed with a solid system definition, the data to describe and illustrate the system, and the insight gained on the time-dynamic nature of the system, the analyst can play a critical role in addressing these challenges.
An underlying theme in this document is the importance of limiting the isolation of transportation analysts. Analyst can bring valuable insights to the overarching process of systems management. With better integration, analytical needs are better defined, better analytical projects are defined, and more insight are gained per dollar spent.
This guidance provides an analytical project scoping cycle to connect analysis results and system characteristics and diagnostics. Both the transportation decision-makers and the data analysts must fully understand the nature of system performance issues and participate in identifying and refining analytic project concepts. In order to use data and transportation analyses as an integral part of a continuous system improvement process, analysts must play an integral role in systems management. They can be engaged in identifying the nature of the problem and possible root causes and then conduct analyses that return insight to the decision to change practices or invest in longer-term solutions. Agencies that engage in a continuous, modern, data-driven analytic project scoping process can expect to see fewer redundant analytical projects, fewer failed analytical projects, and improved decision-making fueled by the insights from relevant and targeted analytical efforts.
This guidance document describes a four-element CIP best deployed as a complete process with all four modules. In general, the authors suggest that readers consider the guidance in typical serial order, that is, reading the introduction first, then the four modules in order. If readers want to refer to this guidance for current projects that are already in latter stages (e.g., data analysis or execution), we suggest reading the introduction before moving to the relevant module to gain insights for these projects. The strongest use of this guidance is not for a single project—but rather for managing a portfolio of analytic projects over time and leveraging past data and models in developing, scoping, designing, and executing future projects.
United States Department of Transportation - Federal Highway Administration