Analysis, Modeling, and Simulation for Traffic Incident Management Applications

Synthesis of Incident Analysis, Modeling, and Simulation Methods

This section describes TIM AMS methods and related applications. In each of the following subsections the current state of the practice is documented. The state of the practice (what is in current use) is discussed separately from the state of the art (that what has been researched).

Survey of Practitioners on TIM AMS Methods

A TIM AMS survey was sent to state departments of transportation (DOT) and metropolitan planning organizations (MPO) in August 2011. The purpose of this survey was to determine current practices in TIM AMS, and to identify areas where practitioners felt that additional guidance would be valuable. The survey questionnaire included eight questions related to TIM:

1. Have you ever conducted a study of the incident impacts on congestion (e.g., delay due to incidents)? If so, please attach the relevant study in an e-mail.
2. Have you ever conducted a study of secondary crashes due to incidents? If so, please attach the relevant study in an e-mail.
3. Do you routinely measure and report secondary crashes?
4. What software tools have you either developed or used to estimate incident congestion impacts or secondary crashes?
5. In what applications that use incident data does your agency currently engage?
6. For which applications would technical guidance be most helpful to you?
7. What kind of information/data would be helpful to you in “making the case” for incident management programs internally with your agency?
8. In terms of your needs for incident information, what types of technical guidance would help you the most?

Eleven agencies responded by September 2011; they are:

• Delaware Valley Regional Planning Commission (DVRPC);
• Florida DOT, District 6;
• Indiana DOT;
• Kansas City Scout (Kansas DOT and Missouri DOT);
• Maryland State Highway Administration;
• Missouri DOT;
• New Hampshire DOT;
• Regional Transportation Commission (RTC) of Southern Nevada;
• Rhode Island DOT;
• Southeast Michigan Council of Governments; and
• Washington State DOT.

Study of Incident Impacts on Congestion

Out of the 11 agencies who responded the survey, 5 (or 45 percent) indicated that they have conducted studies of the incident impacts on congestion (e.g., delay due to incidents).

Study of Secondary Crashes

Out of the 11 response, 4 (or 36 percent) indicated that they have conducted studies of secondary crashes due to incidents.

Measuring and Reporting Secondary Crashes

Out of the 11 response, 5 (or 45 percent) indicated that they routinely measure and report secondary crashes.

Software Tools for Estimating Incident Congestion Impacts/Secondary Crashes

Out of the 11 response, 8 (or 73 percent) indicated that they have either developed or used software tools to estimate incident congestion impacts or secondary crashes.

Use of Incident Data

Figure 2 shows the applications of incident data by the surveyed agencies. The top five applications are:

• Analysis and evaluation of TIM strategies such as use of service patrols;
• Real-time traveler information dissemination;
• Development and evaluation of TIM plans;
• Agency performance reports; and
• Incident prediction and detection.

Figure 2. Use of Incident Data by Agencies

(Source: Cambridge Systematics, Inc.)

Useful Applications of Technical Guidance

Figure 3 shows the useful applications of a TIM technical guidance (with 1 being the most important for TIM AMS application). The top five areas include:

• Relationship between TIM and overall congestion/travel time reliability;
• Development and evaluation of TIM plans;
• Relationship between TIM and overall safety;
• Safety analysis applications such as secondary crash analysis; and
• Analysis and evaluation of TIM strategies such as use of service patrols.

Figure 3. Useful Applications of Technical Guidance, With “1” Being the Most Important

(Source: Cambridge Systematics, Inc.)

Information/Data Helpful for TIM Programs

When being asked what kind of information/data would be helpful for them in “making the case” for TIM programs internally with their agency, the respondents provided the following answers:

• Reliable systemwide speed data.
• Better information related to the benefits of delay/congestion management through transportation systems management and operations.
• Data for reducing congestion and improving safety and linking it with departments such as Operations, Planning, and Safety and Security.
• Injury information is collected for events in which the Service Patrol responds to within the District. But to quantify how many incidents have been averted because of Service Patrol or notification of an incident due to posting of a Dynamic Message Sign (DMS) or through the state 511 system would be helpful information.
• Benefit/cost data, and how incident management is directly tied to safety performance measures.
• Benefit/cost data; find a way to get public officials to understand benefits so they can communicate to media and traveling public; work with upper management so they can present benefit to their board of elected officials.
• Anything would be helpful in justifying operating and maintaining a TMC.
• What can we use to show the cost of secondary crashes and convince our planners and designers to take into consideration traffic safety/congestion/secondary crashes when they design TIM plans for projects.
• Programs to track incident clearance and closure time that can be integrated with control systems.
• Documented safety and congestion improvement results of having a program in place.

Desired Technical Guidance

Figure 4 shows the types of technical guidance that would be helpful. The top two types are technical methods for predicting incident congestion extent (e.g., delay, reliability) and technical methods for predicting incident duration.

Figure 4. Type of Technical Guidance Desired for Improving TIM AMS

(Source: Cambridge Systematics, Inc.)

TIM AMS Methods

This subsection documents the TIM AMS methods and their applications as revealed through a review of the recent literature as well as through agency contacts. It includes what currently is being used by practitioners and what is available from the research. Five types of approaches to evaluating TIM are described:

1. Methods for measuring incident impacts from field data;
2. Methods for predicting impacts of a single incident;
3. Methods for predicting cumulative incident impacts;
4. Methods for predicting incident duration; and
5. Methods for predicting secondary incidents.

Methods for Measuring Incident Impacts from Field Data

Methods for measuring in the field the impacts of incidents on delay must address three challenges:

• The definition of what constitutes delay;
• Collection of delay data over extended periods of time; and
• The parsing of the observed delay among incidents and various other possible causes of delay.

Delay is often defined as the difference between the actual travel time and the free flow travel time. Field measurement of travel time over extended periods has historically been difficult, so agencies and researchers have resorted to spot speed measurements over extended lengths of the facility to compute approximate delays.

State of the Practice

Congestion Monitoring Systems

Congestion monitoring systems and programs are either in place or being developed in the major urban areas of the U.S. Monitoring is done using permanent spot speed measurement stations on freeways, targeted field measurements using floating cars of specific facilities, and/or the use of GPS/cell phone tracking devices by commercial vendors of real-time congestion data. However, the assignment of causality to the measured congestion, and the attribution of delay to incidents are extremely rare in current practice. Assignment of causality is more often done as part of specific research efforts.

One example of how an agency defines delay over extended periods is Caltrans. (Caltrans, Mobility Performance Report 2009, February 2011.) Caltrans defines two delay values: the difference between the observed spot speed and 60 mph (free-flow delay), and the difference between the observed spot speed and 35 mph (breakdown delay). The two definitions of delay are used because the agency’s goal is to minimize breakdown delay.

Automated permanent vehicle detector stations approximately a half-mile apart on urban freeways are used by Caltrans to measure five-minute average spot speeds (24 hours per day, 7 days per week). The vehicle-hours of delay measured at each station is the actual volume measured at the station multiplied by the difference in travel times between detector stations at the actual speed and the delay threshold speed (either 35 mph or 60 mph):

Delay = Volume * [(Length/actual speed) - (Length/threshold speed)]

Caltrans currently does not parse the observed delay into various causes, but has plans to do so in future editions of its statewide Mobility Performance Reports.

The TTI 2011 Urban Mobility Report is an example of a national congestion monitoring report which tallies total delay associated with all causes of delay, but does not assign responsibility to any specific causes, such as incidents. Delay is measured by tracking vehicle GPS devices and comparing off-peak to peak travel times.

State of the Art

Kwon et al. used quantile regression to apportion the causes of measured congestion between incidents and other causes. (J. Kwon, et al., “Decomposition of Travel Time Reliability into Various Sources: Incidents, Weather, Work Zones, Special Events, and Base Capacity,” 2011 Transportation Research Board Annual Conference, Conference CD-ROM, 2011.) In essence, the maximum likelihood contribution of incidents to measured delay is estimated through least squares regression. No underlying traffic behavior model is required.

Skabardonis et al. used a more legalistic approach to separating out incident-related congestion from other congestion. (A. Skabardonis, K. Petty, and P. Varaiya, “Measuring recurrent and nonrecurrent traffic congestion.” In Proceedings of 82^nd Transportation Research Board Annual Meeting, Washington, D.C., January 2003.) First the incident logs were consulted to identify nonincident days. These became the baseline congestion days. These days were then compared to days with incidents. The difference in delay between incident days and incident-free days was considered to be the delay associated with incidents.

The statistical and legalistic approaches are somewhat unusual. More typical are traffic model-based approaches such as List et al. which used the classical queuing model in New York State DOT’s Congestion Needs Assessment Model and updated look-up tables of key parameters to estimate the amount of congestion on arterial streets that might be attributable to incidents. (George List, John Falcocchio, Kaan Ozbay, Kyriakos Mouskos, Quantifying Non-Recurring Delay on New York City’s Arterial Highways, Region 2 University Transportation Research Center, New York, New York 2008.) This approach (like all traffic model-based approaches) requires demand volumes during the incidents and the estimated capacity of the facility before, during and after the incident is present. The estimated delays produced by all of the incidents over the year are summed to obtain incident delays for the year.

Methods for Predicting Impacts of a Single Incident

There are a variety of general purpose and incident-specific analytical tools for predicting the delay impacts of a given incident on a given facility. Since the incident is “given,” these models do not focus on predicting the time, location, and type of incident. They focus on predicting the consequences. Incident-specific analytical tools also may predict secondary incidents and the duration of the incident.

General purpose tools include traffic simulation models, and the recently published 2010 Highway Capacity Manual (HCM) method for freeways. Both facility-specific and systemwide impacts of individual incidents can be estimated using microsimulation models, and mesoscopic simulation models. The 2010 HCM currently is limited to single facility applications.

Microsimulation models require more data and will tend to produce more precise estimates of delay effects than mesoscopic models. Mesoscopic models require more precise facility design (and operations) data than demand models, and tend to produce more precise estimates of delay effects than demand models.

Some of the models specifically tailored to the evaluation of incidents can predict incident duration and secondary incident occurrence as well as delay. General purpose models require this information as input and cannot predict these parameters.

State of the Practice – General Purpose Traffic Operations Analysis Models

General purpose traffic operations analysis models are “state of the practice” for the estimation of the delay effects of specific incidents with a given duration. They do not predict incident duration or the probability of secondary incidents.

There are numerous instances in the literature of general purpose traffic operations analysis models being used in research and in practice to predict the delay effects of incidents and incident management strategies. General purpose traffic operations analysis models of specific incidents come in two basic types: Highway Capacity Manual and Simulation (micro and mesoscopic). (See Volumes 1 and 2 of the FHWA Traffic Analysis Toolbox.)

State of the Art – Incident-Specific Traffic Operations Analysis Models

Incident-specific traffic operations analysis models are “state of the art,” seeing application primarily in research settings.

The iMiT model is an example of a traffic incident-specific traffic operations analysis model (Khattak). (A. Khattak et al., iMiT: A Tool for Dynamically Predicting Incident Durations, Secondary Incident Occurrence, and Incident Delays, 2012 Transportation Research Board Annual Conference, Conference CD-ROM, 2011.) This tool uses statistical models for incident duration and secondary incident occurrence, and uses a theoretically based deterministic queuing model to estimate associated delays. It has been tested in Hampton Roads, Virginia.

AIMSUN ONLINE is an example of a simulation model designed to support real-time incident management decision-making. (A. Torday, et al., Use of Simulation-Based Forecast for Real Time Traffic Management Decision Support: The Case of the Madrid Traffic Centre, European Transport Conference, 2008.) AIMSUN ONLINE deduces the current traffic status on the streets and the actual demand based on data from permanent detectors. With control plans changing dynamically during the day, AIMSUN ONLINE also reads the current control plan operated at each network intersection. Parallel simulation runs are conducted to assess a variety of possible actions that might be applied in order to improve the network situation compared to the “do nothing” case.

Methods for Predicting Cumulative Incident Impacts

In addition to modeling the effect of a single incident, it also is desirable to know what the cumulative effect of incidents is: this accounts for the variability in incident occurrence and severity that occurs over the course of a year.

The tools for predicting cumulative benefits of incident management fall into three categories:

• Tools that predict the effects of incident management for large systems with minimal details or specifics on incident management methods. These tools are typically sketch planning models.
• Tools that predict the effects of incident management for single facilities with a great deal of detail on the specifics of the incident management methods. These tools are typically microsimulation models but with ongoing advances in Highway Capacity Manual methods, may soon include HCM analysis tools.
• Tools that predict the effects of incident management for multiple facility systems with moderate information on the specifics of the incident management methods. These tools are typically mesoscopic simulators employing dynamic traffic assignment.

Sketch planning models are designed to work at very large geographic scales and forecast the system effects of a variety of traveler information, demand management, capacity, and operational improvements, including incident management.

Mesoscopic models, HCM methods, and microsimulation models are generally used to predict the impacts of specific incidents, but when combined with scenario generators and applied systematically to a variety of possible incident scenarios, these more computationally intensive tools can produce predictions of the cumulative benefits of incident management.

More information on these types of tools can be found in Volumes 1 and 2 of the FHWA Traffic Analysis Toolbox.

Sketch Planning Tools

Sketch planning models such as HERS IDAS, and TOPS-BC are designed to work at very large geographic scales and forecast the system effects of a variety of traveler information, demand management, capacity, and operational improvements, including incident management. They apply average incident frequencies, average incident durations, and relatively simple speed-flow relationships to estimate systemwide, long-term effects of incidents on system demand and system delay.

HERS

The Highway Economic Requirements System (HERS) is a model for determining optimal highway investment programs and policies. Two versions of HERS exist, one for national-level analyses, the other, HERS-ST is targeted to state DOT-level analyses. HERS does not have a built in network traffic operations analysis module. This information must be provided to HERS from a separate model, such as a travel demand model network or a mesoscopic model network. The traffic operations effects of different investments in incident management programs are modeled off-line and the results input into HERS.

While numerous operations strategies are available to highway agencies, a limited number are now considered in HERS (based on the availability of suitable data and empirical impact relationships). The types of strategies analyzed can be grouped into four categories: arterial management, freeway management, incident management, and travel information. (Highway Investment Analysis Methodology.) For incident management, HERS can evaluate the following strategies for freeways only:

• Incident detection (free cell phone call number and detection algorithms);
• Incident verification (surveillance cameras); and
• Incident response (on-call service patrols).

HERS was used to model incident management effects for FHWA’s 2008 Status of the Nation’s Highways, Bridges, and Transit: Conditions and Performance Report.

More information on HERS.

HERS: The Oregon Experience (U.S. DOT-FHWA Transportation Asset Management Case Studies: Highway Economic Requirements System: The Oregon Experience.)

In 1999, the State of Oregon developed a customized version of the FHWA Highway Economic Requirements System (HERS) tool for use in conducting investment analysis in the State. The State developed the tool so it could develop more credible estimates of user costs and benefits from transportation improvements. When controversy arose in quantifying the costs of delay in a high-profile incident on one of the state highways, the State developed additional postprocessors to produce estimates of “Unexpected Delay” and “Cost of Unexpected Delay” from the HERS-OR tool.

In a major incident that closed a portion of I-5 for 13 hours, a local newspaper cited estimates of user costs that did not match official ODOT estimates from the HERS-OR model. ODOT took the opportunity to develop additional postprocessors to HERS-OR that produced an Unexpected Delay Map and the Cost of Expected Delay that were acceptable and consistent. This information was shared with all departments and the public and now provides a single consistent source for quantifying delay within ODOT.

IDAS

The ITS Deployment Analysis System (IDAS) is software that can be used in planning for Intelligent Transportation System (ITS) deployments. State, regional, and local planners can use IDAS to estimate the benefits and costs of ITS investments – which are either alternatives to or enhancements of traditional highway and transit infrastructure. IDAS can predict relative costs and benefits for more than 60 types of ITS investments. The IDAS ITS components deployed include:

• Incident detection only; and
• Both incident detection and incident response.

IDAS was utilized to model incident management system in Hampton Roads, Virginia. (2008 Conditions and Performance Report.) The results showed a substantial decrease (9 to 14 percent reductions) in the daily emissions for hydrocarbons (HC) and NO_x in the region for the two options containing incident management improvements when compared with the control alternative without any improvements.

More information on IDAS.

TOPS-BC

FHWA’s Office of Operations sponsored this project to provide guidance on conducting benefit/cost analysis for operations projects, including incident management. The project developed an Operations Benefit/Cost Analysis Desk Reference as well as software to implement it (TOPS-BC). The benefits of operations projects include those related to travel time reliability. For the impacts of incident management strategies, TOPS-BC uses the IDAS procedures.

SSP-BC

Objective:

The SSP-BC was developed for the I-95 Corridor Coalition and FHWA to fill the need for a comprehensive, cost-effective, and standardized Benefit/Cost (B/C) ratio estimation methodology to facilitate evaluation of existing Service Safety Patrol (SSP) programs throughout the country. The tool is based on commonly accepted assumptions and uses an updateable monetary conversion process. A major strength of the tool is not only its utility for evaluating existing programs, but also its applicability in testing numerous what-if scenarios, including the introduction of a new program or the impact of improvements in service response times.

Methodology:

Data in tables used in the tool were derived directly from simulation run results (travel delays, fuel consumption), regression-based estimates (fuel consumption), a novel hybrid statistical-simulation data methodology with improved model fitness (travel delay), computations (emissions, secondary incidents), and from publically available sources (wages, fuel costs, traffic composition, and monetary conversion rates). Power-based equations that incorporate vehicle characteristics and modal parameters (vehicle mass, velocity, and acceleration) in computing instantaneous power demand for each vehicle type category are used in the estimation of fuel consumption and emissions produced (carbon dioxide (CO₂), carbon monoxide (CO), methane (CH₄), Nitrogen Oxides (NO_x), and Sulfur Oxides (SO_x)). The equations are responsive to roadway geometry, traffic volume, grade, and other characteristics of the traffic environment. The output of the tool includes the B/C ratio incorporating user-specified benefit measures, savings in travel delay in vehicle-hours, fuel consumed by passenger cars and light-duty vehicles in gallons, number of prevented secondary incidents, and emission pollutants in metric tons. (E. Miller-Hooks, M. TariVerdi, and X. Zhang, Standardizing and Simplifying Safety Service Patrol Benefit-Cost Ratio Estimation: SSP-BC Tool Development Methodology, Technical Report, I-95 Corridor Coalition, 2012.)

Example Applications

Evaluation of Emissions Impacts of an Incident Management System in Hampton Roads, Virginia

The Hampton Roads Planning District Commission (HRPDC) MPO had invested in deploying ITS technologies in the Hampton Road region. They believed the reduced incidents due to the incident management put in place during the ITS program should logically lead to reductions in emissions. They wanted a tool that could be used to estimate and quantify the emissions reduction. The IDAS software was selected to conduct analysis to quantify the expected emissions reduction. The HDRPC also was interested in quantifying any emissions reductions to the EPA, FHWA, and Air Quality Bureaus to be used in determining the region’s air quality conformity status.

Output runs for a base case and two 2021 scenarios from the regional travel demand model were fed into the IDAS tool for analysis. The IDAS tool was used to estimate the daily incremental emission impacts.

The analysis results “showed a substantial decrease in the daily emissions for hydrocarbons (HC) and NO_x in the region for the two options containing incident management improvement when compared with the control alternative.” Run 1 was the current and near-term ITS deployments, while Run 2 represented greater regional incident management, planned for the future. Though EPA did not ultimately use the results it was a first step for HRPDC to begin quantifying the benefits of their incident management program.

CHART

The Coordinated Highways Action Response Team (CHART) is a joint effort of the Maryland Department of Transportation, Maryland Transportation Authority, and the Maryland State Police. Its mission is to improve real-time operations of Maryland’s highway system through teamwork and technology. From February 2001, all incident requests for emergency assistance have been recorded in the CHART information system and this has significantly enriched the available incident data.

The University of Maryland, as part of the ongoing CHART evaluations, developed a predictive equation model based on running experiments with microscopic simulation:

Excess Delay Due to Incidents = e(-10.19 * (V)^2.8 * (NLB/TNL)^1.4 * (ID)^1.78)

Where: TNL = Total number of lanes;

NLB = Number of lanes blocked;

V = Traffic volume; and

ID = Incident duration.

Using this model, it was determined in 2009 that the CHART program reduced delays by 32.43 million vehicle-hours.

More information on CHART.

Scenario-Based Modeling Approaches

Scenario-based modeling approaches apply mesoscopic models, HCM methods, and microsimulation models repeatedly to a variety of possible incident conditions to arrive at an assessment of the cumulative effects of incident management. There are three primary examples of this approach: The Integrated Corridor Management (ICM) analysis, the FHWA ATDM Evaluation Guide, and the SHRP 2-L08 Reliability in HCM project.

Integrated Corridor Management Analyses

The evaluations of Integrated Corridor Management strategies for Minneapolis, Dallas, and San Diego used travel demand models, mesoscopic simulation models and microsimulation analysis in combination with selected incident scenarios (Table 1). The analysis of multiple incident and weather scenarios (where a variety of incident types can occur at multiple times and locations) were strictly limited to manage analysis costs.

The analysis found significantly positive benefit/cost ratios for integrated corridor management strategies, which include incident management.

This analysis approach can evaluate a wide variety of incident management strategies but it requires a significant investment in analysis effort for the various models that must be employed.

More information on the ICM.

Table 1. ICM TIM Modeling Tools

Model Type	Minneapolis	Dallas	San Diego
Regional Travel Demand Model	Metro model in TP+	NTCOG model, TransCAD	TransCAD
Mesoscopic Simulation Model	Dynus-T – supported by University of Arizona	DIRECT – supported by Southern Methodist University	None
Microscopic Simulation Model	None	None	Transmodeler/Micro

Source: Adapted from: ITS Research Success Stories.

FHWA ATDM Evaluation Guide

The FHWA Active Transportation and Demand Management Evaluation Guide (ATDM Guide) and future replacement for Chapter 35 of the 2010 Highway Capacity Manual currently is under preparation. The ATDM Guide recommends the creation of three different prototypical incident scenarios (no incident, one lane blocked, two lanes blocked) for each of good weather and bad weather days for a total of 6 capacity scenarios. The 6 capacity scenarios are each matched with 5 different levels of demand. The result is 30 scenarios for evaluating incident management and other ATDM strategies.

Special demand, capacity and speed adjustment factors currently are being developed to reflect the effects of incidents and various incident management strategies on these factors. The method will be sensitive to traveler information strategies, speed control strategies (VSL), and lane management strategies (temporary shoulder lane use, etc.).

Once the scenarios have been created and the demand/capacity/speed adjustment factors computed, conventional HCM methods are then used to evaluate facility performance.

The HCM predicted performance for each scenario is weighted by the probability of the scenario occurring over the course of a year to obtain average, median, and any desired percentile (e.g., 95^th percentile) result.

Software to implement aspects of the ATDM methodology is being developed. The proposed ATDM evaluation methodology is being tested on the I-15 corridor in San Diego.

SHRP 2 Projects Relevant for Incident AMS

Several completed and ongoing SHRP 2 projects deal specifically with the prediction of travel time reliability, of which incident impacts are a major component. These SHRP 2 projects are discussed below.

More information on this methodology can be obtained from SHRP 2 staff.

SHRP 2-L08 Incorporation of Reliability in the HCM

The SHRP 2-L08 Reliability Analysis Guide for the Highway Capacity Manual (Reliability Guide) currently is under preparation. The Reliability Guide will recommend the creation of several hundred to several thousand demand, weather, and incident scenarios to predict future travel time reliability distribution.

Two methods for generating scenarios are being considered. One enumerates all possible scenarios and selects the ones of most interest for more extensive evaluation. The other method uses a Monte Carlo approach to generate the scenarios.

Special capacity and speed adjustment factors currently are being developed to reflect the effects of incidents (but not incident management strategies) on these factors. The SHRP 2-L08 project is focusing on predicting existing and future reliability under existing control conditions, rather than predicting how changes in operational strategies can affect reliability.

Once the scenarios have been created and the demand/capacity/speed adjustment factors computed, conventional Highway Capacity Manual (HCM) methods are then used to evaluate facility performance.

An improved HCM Urban Streets method is being developed to better support reliability analysis on arterial streets. The improved method will be able to account for the impacts of queues on upstream signal operation.

The HCM predicted performance for each scenario is weighted by the probability of the scenario occurring over the course of a year to obtain average, median, and any desired percentile (e.g., 95^th percentile) result.

Software to implement the methodology is being developed. The proposed methodology will be tested on a half dozen freeway and urban street data sets.

SHRP 2 L03 Analytical Procedures for Determining the Impacts of Reliability Mitigation Strategies

SHRP 2 Project L03 produced two types of statistical equations based on empirical data for predicting reliability measures. The first set relates the mean congestion condition, as measured by the travel time index (TTI) to a variety of reliability metrics. A strong correlation was found between the mean and the rest of the reliability metrics used, including standard deviation, upper percentiles of the travel time distribution, and on-time measures. The second set relates reliability metrics to demand, capacity, incident blockage, and weather. Publication of the report is expected in late 2012.

Methods for Predicting Incident Duration

When an incident occurs, the timely estimate of its duration plays a key role in the overall incident management process. Reliable incident duration predictions can help traffic managers in providing correct and essential information to road users, applying appropriate traffic control measures at or near the incident location and evaluating the effectiveness of the incident management strategies implemented. The duration of an incident can have several definitions, depending on what is chosen for the start and end times of the incident. Generally the start time is when the incident is first detected by incident management personnel. Ideally the start time would be the time when the incident actually occurred, but this cannot be known with certainty. However, in urban areas, the time between actual start and detection is very small because most incidents are reported by travelers via cell phones within a very short time of the actual occurrence. The end time is usually selected as the time when all lanes are open to traffic or when the last responder has left the scene.

State of the Practice

In Maryland, a Rule-Based Tree Model (RBTM) was applied to develop the prediction model for freeway incident duration by Kim et al. (W. Kim, S. Natarajan, and G. Chang, Analysis of Freeway Incident Duration for ATIS Applications, 15^th World Congress on Intelligent Transport Systems and ITS America’s Annual Meeting, 2008.) The model was developed based on the Maryland State Highway (MDSHA) incident database. The overall confidence for the estimated model was over 80 percent. In cases where RBTM did not provide incident duration within a desirable range, a discrete choice model was developed as a supplemental model.

State of the Art

There are several methods that have been developed in predicting incident duration, as listed below:

• Regression model;
• Hazard-based duration regression model;
• Log-logistic model;
• Prediction/decision tree model;
• Support/relevance vector machine model; and
• Bayesian network model.

Methods for Predicting Secondary Crashes

Secondary crashes are associated with vehicles in close proximity due to a queue formed from a primary incident, the abrupt “end-of-queue” condition caused by a primary incident, collisions with emergency vehicles and personnel, and rubbernecking in both the current and opposite directions of travel. Secondary crashes can be severe, especially at night when visibility is reduced and traffic queues are unexpected. Modeling methods to predict secondary crashes would be greatly enhanced if traffic management center personnel could flag crashes that occur in the queue caused by the primary incident or from opposite direction rubbernecking. Currently, researchers must derive these items analytically.

State of the Practice

Researchers at the Virginia Center for Transportation Innovation and Research also developed a dynamic queue-based tool to identify primary incidents (Secondary Incident Identification tool – SiT). They have used SiT and iMiT together to begin to improve the state of the art in modeling secondary incidents. (A. J. Khattak, X. Wang, H. Zhang, and M. Cetin. Primary and Secondary Incident Management: Predicting Durations in Real Time. Final Report VCTIR 11-R11, Virginal Center for Transportation Innovation and Research, April 2011.)

State of the Art

The following is a list of methods that have been developed and used for quantifying the occurrence and characteristics of secondary crashes.

• Regression model;
• Ordered logit model;
• Probit model;
• Logistic regression model;
• Bayesian network model; and
• Simulation-based secondary incident filtering method.

For instance, a study conducted by Zhan et al. used a comprehensive incident database on I-95 from District 4 of the Florida DOT to identify freeway secondary crashes and their contributing factors. (C. Zhan, A. Gan, and M. A. Hadi. Identifying Secondary Crashes and Their Contributing Factors, In Transportation Research Record: Journal of the Transportation Research Board 2102, 2009.) A method based on a cumulative arrival and departure traffic delay model was developed to estimate the maximum queue length and the associated queue recovery time for incidents with lane blockages.

Vlahogianni et al. also recently utilized neural networks and statistical approaches to study the impact of weather on secondary crashes. (E. I. Vlahogianni, Matthew Karlafits and Foteini Orfanuo., Modeling the Effects of Weather on the Risk of Secondary Incidents, 2012 Transportation Research Board Annual Conference, Conference CD-ROM, 2012.) Their findings were that speed, volume, number of blocked lanes and vehicles involved in crash significantly influence the probability of a having a secondary incident.

A compendium of how transportation agencies are dealing with secondary incidents can be found in the document: Traffic Incident Management Performance Metric Adoption Campaign.

Predicting Incident Characteristics (Independent Variables)

Incident models are built using indicators of incident performance as the predictor variables (e.g., incident duration, lane-hours lost due to incidents). Knowing how TIM strategies affect these independent variables is therefore of utmost importance. A number of studies have been done over the past two decades that can be used for this purpose. SHRP 2 L03 assembled the most recent studies in this area.

You may need the Adobe® Reader® to view the PDFs on this page.