Pseudo-dynamic travel model application to assess traveler information

Transportation 29: 307–319, 2002  2002 Kluwer Academic Publishers. Printed in the Netherlands.

Pseudo-dynamic travel model application to assess traveler information MICHAEL D. ANDERSON1 & REGINALD R. SOULEYRETTE2 1

Civil Engineering at the University of Alabama in Huntsville, Huntsville, AL 35899, USA (E-mail: [email protected]); 2Civil Engineering at Iowa State University, Ames, IA 50010, USA (E-mail: [email protected])

Key words: ITS assessment modeling, traffic modeling, traffic simulation Abstract. This paper reports an effort to estimate potential benefits of Advanced Traveler Information Systems (ATIS) by combing regional travel demand and microscopic simulation models. The approach incorporates dynamic features not yet available in the commercial software market. The suggested technique employs data that are readily available to most urban planning organizations, and is straightforward in its application. The key reported measure of effectiveness is corridor and local system delay, and is sensitive to both the level of penetration of traveler information and the pre-trip and en-route choices drivers make based on this information. The technique is demonstrated on an urban freeway corridor in a medium sized mid-west city.

Introduction Advanced Traveler Information Systems (ATIS) are a major component of the services planned for the national deployment of Intelligent Transportation Systems. ATIS can provide travelers with information regarding conditions of the highways they plan to traverse during a future or current trip. Information for pre-trip planning might include the location and severity of highway incidents, highway construction, delays due to congestion, and weather conditions. Travelers could then adjust departure times and routes accordingly. Similarly, information provided post-departure may be used to make en-route adjustments, though clearly, chances become more limited as the trip proceeds. Two types of models are generally available that may provide tools for assessing the benefits of providing traveler information. Urban planning models (demand models) are regional in scope and assign trips by optimizing routing based on derived destination choices and network congestion. They do not typically, model the impacts of local network detail and many are developed as daily travel models capable of addressing only aggregate, system wide or regional transportation investment or major land-use changes. Traffic microsimulation models, on the other hand, typically include much more detail over a smaller portion of an urbanized area. Traffic patterns are taken as input to

308 the simulation. A combination of the two types of models would seem to hold promise for evaluating ATIS; as the modeling system must both be able to change demand patterns over large extents (the demand model) and be able to focus on the local impact of incidents and the resulting impact of changed demand patterns on a local, near real-time conditions (the traffic model).

Background Sequential methods, even if developed with feedback loops and intersection penalties, lack the ability to identify individual vehicle locations and provide time dependent measures of effectiveness (Miller & Storm 1996; JHK and Associates 1992). Although conventional models can provide daily, or even hourly, traffic forecasts by estimating origin-destination tables and applying pre-defined assignment methodologies, they can not model time slices small enough to respond changes in trip making due to the provisions of real-time traveler information (De Romph 1994; Janson 1991). Microscopic simulation packages, however, estimate vehicle locations at specified times. Many microscopic simulation packages can dynamically route travelers between origin-destination pairs and through the network. However, these packages are incapable of supporting: • dynamic destination choice and • scenarios where traveler information is received by only a subset of simulated travelers (Van Aerde 1997; Kaman 1996). Work is being done developing models that should be able to dynamically assign and microsimulate traffic condition. Ran et al. (1997) describes an analytical path-based multi-class dynamic traffic assignment model in which driver knowledge about the simplified network and ability to modify route choice decisions could be used to evaluate network infrastructure decisions. Janson (1991) developed a dynamic traffic assignment solution, called DTA, based on dynamic user equilibrium, when “all paths between a given pair of zones used by trips departing in a given interval must have equal travel impedances and all paths between a given pair of zones not used by trips departing in a given time interval cannot have lower travel impedance”. Wie (1988) presented a dynamic traffic assignment model for one origin, one destination, and N parallel paths between the two nodes. Lasdon and Luo (1994) studied the dynamic traffic assignment problem using a linear program with concave link exit functions. However, even with all the work currently being performed on dynamic models, they are still in the developmental stage and are not being used in

309 practice to assess traveler information (TRANSIMS 1997). These models incorporate time dependent characteristics into a regional travel model, which could perform an assessment of traveler information. Computationally intensive models and data hungry, they may be out of reach for small to medium sized planning communities for years to come.

Approach We suggest, as a stop gap measure, an approach that approximates a dynamic model by combining the best features of regional travel demand and traffic micro-simulation models. A geographic information system (GIS) is used to manage, store, and graphically present input and output data. The demand model is used to distribute trips and develop network travel patterns. Utility programs were written to convert network geometry and traffic origins and destination (O/D) from the demand model into the format required by the traffic model. The traffic model, by incorporating intersection control effects and other network detail for the local area of interest, calculates “improved” estimates of link travel times, which are then be fed back into the regional model via the GIS interface. During the second and subsequent iterations, the demand model recalculates O/D based upon improved estimates of link congestion coming out of the traffic model. Convergence is assured by the use of a damping factor. Finally, the traffic model is used to generate measures of effectiveness to evaluate the impact of ATIS. Figure 1 outlines this process. This “combined” approach improves on the use of demand models as it assesses the impact of time-dependent incidents. It improves on traffic models as it estimates feedback effects on regional traffic flow patterns. The approach is readily available to current users, and requires only nominal data and computational resources.

Application The integrated travel demand and simulation approach, is demonstrated as a case study for Des Moines, Iowa. A peak period Tranplan model was developed based on an existing 24-hour regional model. Due to network size constraints in the simulation package used the network is then “thinned out” and exported to CORSIM to provide microscopic network assessment capability. The simulation generates more accurate link travel-time estimates, which are in turn used to update the travel demand model using the method of successive averages to smooth travel time updates. The process continues until iterative changes in link assignments fall below a specified tolerance.

310

Figure 1. Flowchart of the combined methodology.

311 At this point, network level measures of effectiveness are calculated by CORSIM. The existing regional model for Des Moines is a 24-hour, Tranplan model with 2378 nodes and 3385 links. The link-node network for the Des Moines MPO area is shown in Figure 2. The model contains 643 traffic analysis zones with daily productions estimated using a cross-classification model and attractions estimated using regression equations. Five trip purposes are used for the model and a gravity model is used to perform trip distribution. For the dynamic assignment, we require an hourly, or shorter, modeling duration, to permit sensitivity to short duration incidents (lasting less than a day). The conversion to an hourly model is performed by applying trip distribution rates borrowed from the literature (FHWA 1998). The regional network is then exported to MapInfo GIS where a sketchlevel network (regional model with limited detail) is defined to include links corresponding to facilities considered when deploying ITS infrastructure in the Des Moines Metropolitan Area ITS Strategic Plan (CTRE 1997). Additional links are included to provide continuity and densify the network in the area

Figure 2. Original Des Moines regional model.

312 of interest (Interstate 235 and the Central Business District). Zone centroids, centroid connectors and nodes are selected to comprise the sketch-level network, which contains 71 zones (aggregated from original 643 zones), 125 nodes, and 232 links. The link-node map of the sketch-level network is shown in Figure 3. As in the original network, the sketch-level network models interchanges as uncontrolled, at-grade intersections. Although some improvements may be afforded by modeling these interchanges with multiple links (ramps, etc.), they are represented as single point intersections in the sketch model to conserve links (CORSIM limits identified earlier). Peak hour link capacities are derived from 24-hour capacities provided in the original Tranplan model through a percentage factor borrowed from the literature (FHWA 1998). Free-flow travel speeds are input directly from regional model link attributes and the GIS program is used to calculate segment length. See (Anderson 1998) for more details on the development of the model and operation of the methodology. The peak-hour sketch level network is exported through MapInfo to CORSIM. Intersection traffic control parameters are input to the simulation control file for all traffic signal locations. For the purposes of this example,

Figure 3. Sketch-level network for Des Moines Metropolitan Area.

313 green-time for each signal is evenly split among approaches, but signal timing splits can be adjusted manually to better simulate actual operations. Traffic volumes for each centroid are set to follow a uniform distribution during the one-hour simulation period, which can also be adjusted to more closely replicate actual conditions and to introduce a more dynamic response. CORSIM generates updated travel times based on congestion levels and intersection traffic control effects. These travel times are input to the sketch model, maintained in MapInfo, to improve the assignment. A modified method of successive averages is used for adjusting travel times. The format for this model is: tn = tn–1 + (1/n) * (tn – tn–1) where: tn is the travel time for iteration n, tn–1 is the travel time for iteration n – 1, and n is the iteration number. This process is continued until the assignment reaches a pre-defined convergence point, obtained when traffic volumes between iterations change by less than 100 vehicles per hour. Several scenarios were developed, each intended to represent travelers with different levels of knowledge of congestion or roadway accident/construction locations. For example, one scenario might represent travelers with no knowledge of an accident on the Interstate Highway 235 (I-235), which has greatly increased travel times from the western suburbs into downtown, and all proceed on a work trips using I-235. Another level of knowledge might allow for travelers to know about the incident and divert from I-235 to a surface street. These scenarios are then combined, where an assumed portion of the population has knowledge of delays and will divert and the other portion does not divert. The combined scenario is assigned and simulated in CORSIM to generate relevant measures of effectiveness. Comparisons are made by changing the level of traveler information (pre-trip versus en-route), the percentage of travelers who receive the information and divert, and the delay following the incident before the information is available. Pre-trip information system assessment To assess the benefit of pre-trip traveler information, the baseline regional network is modified to include a major construction project on I-235 through downtown Des Moines. During construction, the facility is modeled with one of three lanes closed, and the speed limit reduced from 55 mph to 40 mph.

314 Further, a parallel major route, normally considered to be a preferred detour, is also under construction. Two variants of the regional baseline model are developed, one with arterial construction and one without (both versions include the reduced freeway capacity). In the construction variant, increased travel time on the arterial results in a second origin-destination table for trip purposes other than work trips. This reflects that some drivers may change destinations due to traffic conditions – a limited form of dynamic distribution, in which modeled trips have different destination locations in the two scenarios. However, the homebased work portion of the trip table is held constant (it is assumed that these trips have fixed OD patterns on a day-to-day basis). Following the procedure described above, sketch level models are prepared and simulated for each “level” of traveler information (all drivers have information and all drivers do not have information). Drivers who use the information and divert are assumed to vary from 5 to 25 percent of the total. Travel patterns in the two sketch-level models are combined accordingly, e.g. for 5 percent traveler information compliance, traffic assignments are comprised as 5 percent load from the “with information” model and 95 percent load from the “without information” model. In the absence of arterial construction, the sketch level dynamic model estimates 16,700 person-minutes of delay during the peak hour. With arterial construction, and without providing any traveler information, the delay estimate increases to 21,600 person-minutes. As the percentage of travelers with information increases from 5 to 25%, delay is reduced. Benefits from travel-time savings are calculated using a $10.30 per-hour, per-vehicle value of time (9). For the scenario tested above, benefits of traveler information for one hour are listed in Table 1. The minimum capital cost for pre-trip information is reported to be around $15,000 for a simple Internet site or cable TV access (9). Amortized over 3 years at a discount rate of 7 percent, annual capital cost for the system is $5,200. Operations and maintenance costs will vary depending on how traffic Table 1. Benefits of a pre-trip information system, per peak hour. Percent using system

hourly delay (person-minutes)

delay reduction (person-minutes)

delay reduction (person-hours)

cost savings of reduction (dollars)

00 05 10 15 20 25

21,608 21,519 20,902 19,804 18,483 17,653

00– 0089 0706 1804 3125 3955

0– 01.5 11.8 30.1 52.1 65.9

00– 0$15.28 $121.20 $309.69 $536.46 $678.94

315 information is obtained. A simple scenario where data are gathered from FAX and telephone reports from local authorities, the cost of compiling construction information and major incidents on a map could be conducted by an hourly employee (a student work-study) for less than $10,000 per year. Total annualized cost of the system, therefore, may be approximately $15,600. The number of hours the system must be used to justify investment depends on the provision and compliance with traveler information. If only five percent of drivers comply with the pre-trip information system, approximately 1000 annual hours of compliance are required to break even. However, if as many as 25 percent of the drivers comply, the benefits of the system are expected to quickly recover investment costs in as few as 25 hours. En-route information system assessment An assessment of en-route traveler information technologies is performed using the pseudo-dynamic model, again, with convergence attained using the modified method of successive averages. The assessment uses the same network as the pre-trip assessment with the interstate construction. A short duration incident on the arterial replaces the arterial construction activity (essentially, the “full-hour” arterial restriction is shortened to less than one hour). Following the assessment methodology, a certain percentage of the travelers are allowed to alter their route and destination choices based on information about the incident after reaching a post-awareness period. This section presents the results of providing improved traveler information via highway advisory radio system (HAR) and variable message signs (VMS). Benefits and costs are analyzed for each individually, and for a system that uses both technologies. As with the pre-trip scenario, two regional models are developed. One model contains only the interstate construction and the other contains the interstate construction and a traffic incident causing a lane blockage on the arterial (modeled as if the lane were blocked for the entire period). Again, origindestination trip tables are developed for each situation allowing two OD patterns. The difference between the scenarios is that the in the pre-trip assessment, the combined pattern is assigned for the entire hour. In the en-route assessment, the assignment follows the OD pattern where travelers are not provided any information up to a specified time when the incident is identified, only then does the model switch to the combined pattern. A twenty-minute incident is scheduled to begin ten minutes into the simulation. This incident causes a single-lane blockage for traffic entering the downtown area. Before and during the first ten minutes of the incident, all traffic assignment is based on patterns that reflect no knowledge of the incident. During the last 10 minutes of the incident, travel patterns from the

316 two assignments are combined (according to the percentage of drivers expected to receive and comply with the information). After the incident is cleared (30 minutes into the incident), assignment returns to the original, pre-incident pattern. Four scenarios are tested: • The incident with no traveler information (base case), • The incident with five percent of travelers responding to a (HAR), • The incident with 20 percent of travelers responding to a system of (VMS), and • The incident with 25 percent of travelers responding to information provided by a combination of the two technologies. Hourly delay and potential cost-savings for these systems are shown in Table 2. The capital cost for HAR systems range from $55,000 to $110,000 (9). Using a seven-percent discount rate, a 15-year life, and 15% maintenance costs, and $10,000 per year operations cost, the annual cost of the system is between $16,800 and $23,700. Assuming that an incident occurs once during the peak, every other day, the total annual benefit (assuming 5 percent compliance) is $18,800 (again, a $10.30 for the value of time is assumed). This results in benefit-cost ratios ranging from 0.9 to 1.25. Cost estimates for VMS range from $55,000 to $90,000 (arterial) and $115,000 to $190,000 (freeway) (9). It is assumed that four variable message signs are required to notify travelers about this incident. Using two VMS signs on each of the freeway and arterial locations, and again, using a seven percent discount rate, 15 year life, 15 percent maintenance and $10,000 per year operations cost, annual cost ranges between $37,288 and $50,858. Assuming that an incident occurs every other day, the total annual benefit (assuming 20 percent compliance) is $86,086 per year. This results in a benefit-cost ratio ranging from about 1.7 to 2.3. For both technologies, the total cost is assumed to be less than $64,500 (without duplicating operating costs). Using the potential cost savings of $618 per incident, annually, there need to be 104 twenty-minute incidents to break even. Table 2. Hourly delay experienced for the different ITS infrastructure deployment scenarios. Scenario

hourly delay (person-minutes)

hourly delay (vehicle-hours)

time savings (vehicle-hours)

cost savings (dollars)

No technology HAR 5% VMS 20% HAR and VMS

18,509 17,901 15,695 14,924

308 298 262 248

– 10 46 60

– $103 $473 $618

317 Conclusions This study recommends an approximate dynamic modeling approach to assess traveler information technology deployment. The modeling approach integrates a regional travel demand model and micro-simulation package in a GIS environment. A case study was developed for the Des Moines, Iowa metropolitan area and two traveler information scenarios were tested (pretrip and en-route). This study is about the evaluation of new systems. Intelligent transportation systems (ITS) represent a new way to approach the provision of transportation services – services one would not necessarily expect conventional assessment methods to effectively address. While development of new transportation modeling methods has begun (through the Travel Model Improvement Program), practical tools remain beyond the reach of many planning agencies. While no single tool can assess all ITS services, this work shows that, by combining familiar approaches, analysts can effectively address some ITS services (e.g. traveler information). The methodology builds upon the strength of a widely used urban transportation models to assess regional implications of ITS. It also incorporates the graphical and spatial data management strengths of desktop mapping. The approach also draws on the capabilities of CORSIM, FHWA’s traffic microsimulation modeling environment. CORSIM is used to compute time dependent measures of effectiveness such as intersection queues, stop delay, fuel economy and vehicle emissions, while Tranplan permits limited-dynamic destination and route choices and assignments based on optimal as well as perceived-optimal paths. The conceptual approach described is equally applicable to other demand and traffic models. In the case study, the methodology is successfully demonstrated as model information is shared between packages to approximate dynamic assignment. The model makes use of familiar tools and readily available data. Although many simplifying assumptions were made in the case study (e.g. the percentage of travelers who divert due to receiving traveler information), the scenarios assessed in the case study indicate that providing improved traveler information can reduce system travel time and result in positive benefits for system deployment. Based on our assumptions, benefit cost ratios for en-route information systems are shown to range from 1.08 for highway advisory radio to 2.0 for variable message signs. The intended contribution is the demonstration of an approximate dynamic modeling environment suitable for assessing traveler information. The identification and combination of optimal and “perceived-optimal” paths in a regional network simulation is something new. The system allows region-wide assessment of pre-trip and en-route traveler information services and incor-

318 porates travel time feedback into the trip making and route selection processes. The system may provide a practical approach to assessing ITS improvements. In general, conventional transportation modeling environments can be integrated to perform analyses of new transportation systems.

Acknowledgements The authors would like to thank the Center for Transportation Research and Education at Iowa State University, The Federal Highway Administration’s Eisenhower Fellowship Program, the Des Moines Metropolitan Planning Organization and the Iowa Department of Transportation for making this research possible.

References Anderson M (1996) Assessing the Benefit of Providing Improved Traveler Information: A Pseudo-Dynamic Modeling Approach. A Dissertation submitted to Iowa State University, Ames, Iowa. Center for Transportation Research and Education (CTRE) (1997) Des Moines Metropolitan Area ITS Strategic Plan. De Romph E, Grol H & Hamerslag R (1994) Applications of dynamic assignment in Washington, DC, metropolitan area. Transportation Research Record 1443: 100–109. Federal Highway Administration (FHWA) (1978) Quick-response urban travel estimation techniques and transferable parameters. National Cooperative Highway Research Program Report 187. Janson B (1991) Convergent algorithm for dynamic traffic assignment. Transportation Research Record 1328: 69–80. JHK and Associates (1992) Guidelines for Transportation Planning. Kaman Sciences Corporation (1997) TSIS User’s Guide: Corsim User’s Guide Version 1.02 Beta. Lasdon & Luo (1994) Computational experiments with a system optimal dynamic traffic assignment model. Transportation Research Part C 2(2). Miller H & Storm J (1996) GIS design for a simultaneous equilibrium model. Proceedings of the 1996 GIS-T Symposium (pp. 41–79). MO: Kansas City. Ran B, Li I & Soetopo W (1997) An analytical path-based multi-class dynamic traffic assignment model. Transportation Research Record Preprint 970908. TRANSIMS (1997) TRANSIMS Project Description, Travel Model Improvement Program. http://www.bts.gov/tmip/publ/transims.htm. Van Aerde M (1997) Integration Release 2: User’s Guide-Volume I: Fundamental Features. Transportation Research Group, Queen’s University, Kignston, Ontario, Canada. Wie B (1988) An application of optimal control theory to dynamic user equilibrium traffic assignment. Transportation Research Record 1251.

319 About the authors Michael Anderson is an Assistant Professor of Civil and Environmental Engineering at the University of Alabama in Huntsville. He conducts and manages research on urban transportation modeling, simulation of transportation infrastructure, geographical information systems, and rural public transportation. Reginald Souleyrette is Associate Director for Transportation Planning and Information Systems for the Center for Transportation Research and Education and is also an Associate Professor of Civil and Construction Engineering at Iowa State University. He conducts and manages research on the application of geographic information systems to transportation (GIS-T), traffic safety, and transportation planning and modeling.