Analysis of a Traffic Diversion Strategy Using Corism Traffic Micro-Simulation Software

Throughout the United States metropolitan area freeways are operating at or near capacity. Any disturbance in the traffic flow on these freeways, planned or unplanned, can result in varying degrees of congestion. Incident management programs have been established in urban areas nationwide to help reduce the magnitude of incident induced congestion. These programs focus mainly on the incident identification and response stage, and have only recently begun to develop the tools and techniques needed to manage the recovery stage. One technique that has been becoming more widely employed, with the increased focus incident management and overall traffic systems management, is alternate route traffic diversion. However, even when diversion strategies are employed, often only the main line freeway is evaluated and insufficient consideration is given to the effect of the diversion on the alternate route. Also, traffic diversion strategies are typically deployed only during extreme incidents and are seldom deployed or even analyzed for how they would help mitigate congestion related to minor traffic incidents. If delay on the freeway network as a whole is to be minimized, incident management programs need to incorporate comprehensive traffic management strategies and decision aids for the defining of traffic diversion strategies. This study utilized Federal Highway Administration's (FHWA) traffic microsimulation software, CORSIM, to evaluate a freeway route diversion strategy that would increase traveler safety and alleviate congestion caused by minor incidents. The goal of the project was to reduce the impacts of minor incidents on a freeway

identification and response stage, and have only recently begun to develop the tools and techniques needed to manage the recovery stage. One technique that has been becoming more widely employed, with the increased focus incident management and overall traffic systems management, is alternate route traffic diversion. However, even when diversion strategies are employed, often only the main line freeway is evaluated and insufficient consideration is given to the effect of the diversion on the alternate route. Also, traffic diversion strategies are typically deployed only during extreme incidents and are seldom deployed or even analyzed for how they would help mitigate congestion related to minor traffic incidents. If delay on the freeway network as a whole is to be minimized, incident management programs need to incorporate comprehensive traffic management strategies and decision aids for the defining of traffic diversion strategies.
This study utilized Federal Highway Administration's (FHWA) traffic microsimulation software, CORSIM, to evaluate a freeway route diversion strategy that would increase traveler safety and alleviate congestion caused by minor incidents.
The goal of the project was to reduce the impacts of minor incidents on a freeway II system through the use of a methodically analyzed diversion strategy. The four specific objectives of the research were (1) to determine the impact of varying degrees of traffic diversion will have on the network evaluators for varying levels of traffic volumes encountering different incident situations ; (2) to evaluate CORSIM and companion software as tools for performing this analysis; (3) to determine and document the procedure of the use of CORSIM in this facility; and ( 4) to recommend a diversion practice. The research analysis used typical measures of effectiveness to evaluate the effects of minor incidents and traffic diversion on a mainline and alternate route.
The results of the research show that CORSIM is a valuable tool for researchers, planners, or transportation engineers performing diversion strategy analysis. By utilizing CORSIM to simulate specific route diversion strategies valuable insights were gained into the effects that minor unplanned traffic incidents and deployed route diversion strategies will have on average network measures of effectiveness. It was found that the traffic condition the network is experiencing at the time of a minor incident and route diversion has distinct impacts on the network evaluators. CORSIM simulations demonstrated that the four levels of traffic condition tested produce distinct route diversion strategy recommendations. It was shown that the diversion modeled had a negative impact under AM peak traffic conditions, a slightly positive impact under 3/4 AM peak traffic conditions, significantly positive impact under 2/3 AM peak traffic conditions, and no impact under 1/2 AM peak traffic conditions. The significance of these results is that they indicate that it is only beneficial to the entire network to divert traffic for certain incident situations when the lll network is operating at 3/4 AM peak or 2/3 AM peak traffic. At AM peak and 1/2 AM peak traffic condition, diversion was not found to be warranted for any of the minor incident situations modeled. IV

ACKNOWLEDGEMENTS
Before I begin acknowledging all the people who made this research and thesis document possible, I want to thank God for the opportunities and gifts afforded me.   (Garrison and Mannering 1990). A report for the California Department of Transpo11ation (Cal trans) placed the cost of incident-related congestion at approximately $1 million per day (Reiss and Dunn 1991 ). It is crucial to understand that although major traffic incidents cause severe traffic flow disruptions, minor traffic incidents can cause a substantial portion of the total delay attributable to incidents. A report by the Federal Highway Administration (FHW A) stated that minor incidents are responsible for 65% of all incident related delay, with major incidents accounting for the remaining 35% (Reiss and Dunn 1991).
In addition to causmg delays and increasing highway user costs, vehicles subjected to congestion produce excess emissions. Congestion is also being linked to the growing "road-rage" phenomena. Since transportation officials have realized that metropolitan areas will not be able to consistently curb freeway traffic congestion, incident related or otherwise, by adding highway infrastructure, there has been a nation-wide drive towards transportation system optimization. There are many components of transportation system optimization, one of which is incident management.
Incident management programs have been established in urban areas nationwide to aid in the reduction of the magnitude of incident induced congestion.
These programs focus mainly on the incident identification and response stage, and have only recently begun to develop the tools and techniques needed to improve the recovery stage, particularly the utilization of traffic diversion strategies. Even when diversion strategies are employed, often only the main line freeway is evaluated while insufficient consideration is given to the effect of the diversion on the alternate route.
Also, traffic diversion strategies are typically used only for extreme reductions in capacity and are seldom deployed for minor traffic incidents. If delay on a traffic network as a whole is to be minimized, then it is essential that incident management programs incorporate comprehensive traffic management strategies and innovative decision aids for the analysis of traffic diversion strategies so that they can be deployed for any situation that warTants them.

Goals and Objectives of the Research
In the City of Providence, RI, there is a segment of Interstate 95 (l-95) refetTed to as "Thurber's Avenue Curve". Travelers to the downtown area of the City of Providence and points further north often encounter extended delays because of congestion caused by traffic incidents on this curve segn1ent because of substandard 2 freeway geometry and high traffic volumes. The diversion strategy to be investigated utilizes Highway RI Route 10 northbound (Rt. 10 NB) as an alternate route around the "Thurber's A venue Curve" segment described above and shown in Figure 1-1.
Pictures of typical sections ofl-95 NB and Rt. 10 NB in the study area can be seen in Appendix A. This diversion strategy is an excellent test case for this freeway-tohighway diversion study because Rt. 10 operates at or near capacity at peak travel periods and detailed analysis of the effects of the diverted traffic is required before a diversion plan could be deployed. Furthermore, when the RIDOT ITS initiative is complete, traffic diversion for minor freeway incidents will be feasible.
Traffic diversion for minor incidents will be possible because the ITS deployed will include the necessary data collection systems, surveillance systems, information dissemination systems, and central management systems to assess the status of incidents and guide travelers to alternate routes. The specific nature of these systems is explained in Chapter 2. This research could lead to the production of an incident management course of action for operators at the central transportation management center to follow at the time of an incident. With the ITS and this guideline in place route diversion strategies could be deployed efficiently based on sound analysis.

Overview of the Thesis Document
This document begins with a detailed discussion on why the study is significant undertaken and what the study document strives to achieve. This introduction is followed by background information regarding traffic diversion and its place in transportation management today. Following the background information is

Historical Perspective of Traffic Diversion Practice in the United States
Since the beginning of vehicular traffic, motorists have been deviating from mainline routes to alternate routes in the event of roadway congestion caused by traffic incidents. When roadways were few, alternate routes to desired destinations were few.
However, as the automobile quickly replaced horse and buggy and, in many densely populated places, intra-city trolley ways as well, roadways began to blanket urban landscape. Over the years, roadways rapidly increased in length, traffic capacity, complexity, and function. The different types and usage of roadways naturally led to classifications, so that roadway function could be readily identified by classification.
The most common classifications are: land access, a1ierial , collectors, highway, and freeway. The following is an example of how the roadway label relates to the function: a land access roadway is typically a neighborhood road that allows a private landowner to access their land by automobile. With these new vast networks of interwoven roadways, each having different functions, types of controls (signs and signals) and restrictions (speeds), travelers promptly identified various routes, composed of a hybrid of roadway types, to the same destination.
Until 1971, traffic diversion to alternate routes during the time of incidents were the product of either a segment of roadway experiencing a total loss of capacity and an emergency diversion to any possible alternate route or the familiarity of 7 motorists with a given roadway network and their ability to readily navigate the system based on commercial radio traffic reports or learned knowledge and assessment of the prevailing traffic conditions. In 1971, District 7 of the Caltrans pioneered the development and deployment of alternate route plans for responding to the occurrence of major traffic incidents. Major traffic incidents are defined as incidents that severely reduce or eliminate a roadway ' s capacity for an extended period of time (Roper 1991). District 7 initiated the process of developing 2,500 alternate map routes for 475 miles of freeway. Each map identified several key components vital to the alternate route deployment process including identification of the problem location, primary and secondary alternate routes, deployment guidelines, manpower requirements and locations, required signing, necessary closures, responsible parties and associated phone numbers, and special notes unique to the incident area (Dunn et al. 1999). With Caltrans leading the way, other state agencies began to develop route diversion strategies for major incidents, recognizing that time, money, and resources could be saved and overall system safety would be increased.
Traffic diversion is based on the fact that drivers will divert from the usually quickest, or most popular route, to another alternate route if they perceive that they can reach their destination faster and safer using the alternate route. The alternate route is often times only utilized if the preferred route, or mainline route, is operating at a congested level. Congested levels are reached as vehicle demand exceeds a roadway's capacity, or if a segment of that roadway's capacity is reduced by an incident. A traffic incident is any event, planned or unplanned, that reduces the capacity of a segn1ent of roadway effecting upstream traffic conditions (Dunn et al. 8 1999). An unplanned traffic incident is commonly referred to as a " traffic accident" .
A planned traffic incident can take the form of routine roadway construction or some special event that has been pre-determined to cause traffic congestion. Furthermore, traffic congestion is broken into two main categories: recurring and non-recumng congestion (Duru1 et al. 1999). Recurring traffic congestion occurs usually during commuter travel time periods causmg the roadway to consistently experience a demand that exceeds its capacity. Examples of recurring congestion are Los Angeles freeways and Interstate 95 through New York City. Commuters on these roadways expect to experience delay under usual commuter period traffic flow conditions. Nonrecurring congestion occurs when a roadway segment's capacity is suddenly reduced or the roadway's demand exceeds its capacity either by a planned or unplanned incident (Dunn et al. 1999). This study investigates an alternate route strategy that would be deployed in the case of non-recurring congestion stemming from an unplanned traffic incident causing a sudden reduction in freeway capacity.
The practice of diverting traffic only for incidents that result in a complete loss of capacity has been the standard for incident management program traffic diversion plans. Until recently, transportation management centers (TMCs) have had limited means of collecting, processing, and responding to incident infom1ation quickly enough to make it beneficial to divert traffic for incidents that do not block all Janes (Dunn et al. 1999). Also, analysis of diversion strategies for various incident events calculation or with the assistance of computer simulation models has been extremely timely. However, RI and many other states now emplo y advanced transportation management centers (ATMCs) that integrate real-time traffic surveillance with 9 variable message signs (VMSs) and/or dynamic message signs (DMSs) along the freeway. These and other ITS technologies will enable significantly more efficient incident data collection, data processing and information dissemination, making traffic diversion for minor incidents feasible. While ITS will facilitate the implementation and operation of actual traffic diversion, advances in computer processing and traffic simulation software, specifically with regards to the model input process and database construction, are making the use of simulation models increasingly economic and practical.

Current Traffic Diversion Practice in the United States
In 1999 Based on the agencies surveyed m the aforementioned synthesis it can be stated that most deployed traffic diversion strategies responding to unplanned traffic incidents are the product of: • interagency coordination, including identifying the lead agency; • proven traffic incident detection techniques, including police patrols and roadway users ; • specific pre-detem1ined sets of decision criteria based on roadway and traffic incident characteristics; • resources to inform motorists of prevailing traffic conditions, including variable and dynamic message sign and highway advisory radio; and, • resources to guide motorists along the alternate route.
The synthesis relates how various agencies use different criteria in deciding if traffic diversion is watTanted. Also, it relates the different components for detection and guidance along the alternate route and states that all the components listed must be present in some form.

Traffic Diversion Practice in Rhode Island
The State of Rhode Island (RI) is on the cusp of deploying an ITS initiative that will revolutionize roadway system management throughout the state. This initiative already includes an advanced central transportation management center (TMC) in Providence, limited highway advisory radio (HAR) on the AM frequency l l 161 O, four operational closed circuit video surveillance cameras, movable variable message signs (VMS) along I-95, and a volunteer program that uses participating commuters as probes during peak travel periods. With these components in place and many others nearing operation, RI will have the capability to perform efficient and effective route diversion during unplanned incidents.
Presently infom1ation on traffic incidents that may affect RI motorists is instantly sent to the RJDOT TMC to be disseminated appropriately through ITS components and other media.

Introduction to Traffic Simulation
The earliest computer simulation work in highway transportation occurred in the 1950's when the Road Research Library in the United Kingdom undertook an intersection simulation (May 1990). The first simulation work in the US was published in 1953 and repo1ied on intersection and freeway models developed at the University of California at Los Angeles. This initial work was followed by intersection work at the University of Michigan, major arterial simulation at Philco, New York Port Authority, and freeway ramp merging simulation at the Midwest Research Institute. From the l 950's, computer simulation grew rapidly through the 1960's, 1970's, and 1980's (May 1990). However, it was not until the 1990's, and the advances in micro computing, that computer simulations run efficiently on personal computer systems.
Computer simulation models are typically employed to predict how real world systems will behave under a set conditions without having the real world system involved (Elefteriadou et al. 1999). They can play a major role in the analysis and assessment of a complex highway system and its components (May 1990). Computer simulation models incorporate various analytical techniques, such as car-following theory, lane changing theory, capacity analysis and emissions analysis. Simulation models are typically distinguished in the following ways:

13
• Simulation models predict system performance on the basis of a representation of the temporal or spatial interactions between system components (Elefteriadou et al. 1999); • Empirical models predict system performance on the basis of relationships developed through statistical analysis of field data (Elefteriadou et al. 1999); and • Analytical models predict system performance usmg relationships among system components developed through theoretical considerations, tempered, validated, and calibrated by field data. (Elefteriadou et al. 1999) Within traffic simulation there are three distinct categories of how the models replicate traffic: micro-, meso-, and macroscopic. The difference lies in the level of detail at which the traffic flow phenomena are being represented (Elefteriadou et al. 1999). Microscopic models capture the movement of every vehicle and contain processing logic to describe how the vehicles will behave. The behavioral description includes acceleration, deceleration, lance changes, passmg maneuvers, turning movement execution, and headway gap acceptance. Macroscopic models employ flow-rate variables and other more general describers to model the traffic movements.
Mesoscopic models fall in between micro-and macroscopic models. Mesoscopic models typically model the movement of clusters or platoons of vehicles and mcorporate equations that indicate how these clusters interact.
Each of the three categories can be progranuned as either deterministic or stochastic. A model is deterministic if no element of the model is subjected to 14 randomness. The significance of a model being deterministic is that every simulation with the same inputs will produce exactly the same outputs. For a deterministic model, all model parameters are known in advance and all the outcome determinations by the model can be predicted with certainty before the simulation begins Elefteriadou et al. 1999). The model is stochastic if random variables are used during the simulation to determine either specific values for model variables or actions of simulated vehicles (Elefteriadou et al. 1999). A stochastic model will generate varied results based on random number sequences. These are either set as default values in the program, or are user specified. The significance of a model being stochastic in nature is that identical inputs can produce varied outputs, effectively allowing for experimental sampling.

Simulation Software Utilized in Research
FHW A has been utilizing and developing various traffic simulation models to aid in the analysis of these effects. These models are micro-and macro-scopic in logic and combined with other companion software make up FHW A's TRAF family of software. The subsequent sections will describe the TRAF software applied in this research

CORSIM
The microscopic simulation model of the TRAF software is CORSIM, an abbreviation of Corridor Simulation. CO RS IM combines two of FHW A's original microsimulation models, NETSIM, used for modeling urban surface roadway systems, and FRESIM, used for modeling freeway roadway systems. By combining these models, CORSIM is able to simulate more complete and realistic networks of freeways and urban surface roadways ("CORSIM" 1998). Presently, CORSIM is being employed by transp011ation engineers and planners to analyze a wide range of projects. From designing high occupancy vehicle (HOV) lanes to evaluating complex interchanges to analyzing intricate signalization schemes needed for urban intersections, CORSIM software has given traffic and road design engineers valuable information into how their designs will function under various traffic and roadway conditions. Figure 2-1 displays a typical section of a CORSIM output file as seen by the user on a computer monitor. It is important to note that input files are no longer generated in CORSIM. They are created in the graphical user interface (GUI) companion software, ITRAF, which will be described in a subsequent sub-section. to simulate the actions. The next two subsections present discussion and description of the lane-changing and car-following models within FRESIM.

Car-Following Model
The car-following theory programmed in FRESIM is significant because it defines bow vehicles in the model will interact with the environment and the other vehicles on the network. In the FRESIM model, each vehicle in each time increment is assigned one of the following characteristics: "a follower" (a vehicle following another vehicle) or "a leader" (a vehicle with no leader). The encoded car-following theory assumes that a follower vehicle will maintain a safe time-space between itself and its leader. This time-space is given by Equation 1 of the PITT Car-Following model and is presented in Appendix B. When the space is insufficient to maintain a "time-space safety cushion" in order to avoid a collision, the vehicle will decelerate in order to maintain a safe distance. At any given time interval, the acceleration of the follower vehicle is ~etermined by the behavior of the " leader" vehicle and the downstream geometric conditions. This acceleration is compared against the vehicle's performance capabilities and adjusted if necessary. In order to avoid a collision, an emergency constraint overrides the car-following acceleration and maintains the safety cushion. This situation continues until the distance between the blockage and the front bumper of the vehicle is less than or equal to 5 feet. The vehicle's speed and acceleration is then set to zero ("CORSIM" 1998). This situation continues until the distance between the blockage and the front bumper of the vehicle is less than or equal to 5 feet. The vehicle's speed and acceleration is then set to zero ("CORSIM" 1998).

21
3. Vehicle is Approaching the End of an Auxilimy Lane. The behavior of a vehicle approaching the end of an acceleration auxiliary lane is identical to that of the lane-drop behavior as described above ("CORSIM" 1998).
4. Vehicle is Not Affected by Geometrics. Any vehicle that is not influenced by any of the previously described cases will attempt to increase its acceleration to the maximum possible rate in an effort to attain the user specified free-flow speed of the facility, which depends on the freeway geometrics and the vehicle's operation characteristics ("CORSIM" 1998).

Lane-Changing Model
Lane-changing logic determines the amount of risk that a driver of a lanechanging vehicle will accept (lead gap) and the amount of risk that a driver in the target lane will accept (lag gap). Figure 2-  • Advantage -pertains to when the lane change will be advantageous to the driver.

Lead Time Ga
• Urgency -pertains to how strong the desire is to change lanes.

24
A detailed discussion of these behavioral parameters and the governing equations can be in the Appendix B.
3. Anticipatory Lane Changing. Anticipatory lane changing refers to the lane changes that are performed by through moving vehicles to avoid potential slowdown caused by the traffic merging from a downstream on-ramp.

FRESIM Limitations
"Although FRESIM is probably the best freeway simulation program available, it does have some limitations that could be overcome with minor programming changes and enhancements" (Roess and Ulerio 1997). It has been documented that one of the main problems with FRESIM is its inability to predict the merging process with a high degree of accuracy. Another problem that has been encountered is that FRESIM allows some vehicles to miss their exit point. This is a very serious consideration for the research being performed. If vehicles miss the correct exit point, it inaccurately reflects the volume of traffic on both the main and alternate route. Because of this phenomenon the modeling technique was altered from its original fomrnt. The technique employed is described in detail in the research methodology.
Another commonly cited problem with CORSIM is that the output data is not the most relevant. Specifically, some of the summary statistics are reported on a cumulative basis, which in most instances requires the user to perfom1 additional calculations to find the actual statistics for a given time period (Roess and Ulerio 1997). Also, the CORSIM program provides no graphical output for some of the key network statistics.
Finally, and most importantly, much of the programming logic in FRESIM is based on field data that in some cases may not have been as extensive as necessary in the development of a program of its scope (Roess and Ulerio 1997). It has been stated that, "many of the deficiencies of the program could be overcome by conducting additional calibration/validation and sensitivity analyses to refine the models" (Roess  • North Dakota Department of Transportation (NDDOT} -FHW A R&D assisted NDDOT in the use of TSIS and CORSIM to perform advanced operational analysis. The tool was employed to evaluate alternative corridor designs for a complex series of integrated freeway and non-freeway interchanges. CORSIM allowed NDDOT to demonstrate and communicate significant freeway performance degradation due to the introduction of one particular design where freeway speeds dropped approximately 20 mph. These speed differential s were detern1ined to be significant and to pose a serious safety and operational problem. ND DOT realized a cost savings of at least $2.5 million by eliminating the design and construction of a proposed on ramp. 33 Another benefit was that approximately $600,000 annual savings in peak-hour user costs. controllers programmed with the identical signal plans to those existing at the Route 7 intersections, with minor modifications to allow signal preemption. In this carefully controlled hardware-in-the-loop environment, CORSIM provided the microscopic simulation and tabulation of measures of effectiveness, but instead of CORSIM emulating controller features, the simulation package sent detector information to the physical controllers and read back phase indicators. Since CORSIM tabulates performance measures of effectiveness (MOE's), quantitatiye results with and without preemption measurements were obtained.
Results showed that, for the geometric and operational conditions studied, the impact of emergency signal preemption on the signal coordination of the corridor was minor. Although several of the preemption cases had "statistically significant means" when compared to the base case (no preemption), the magnitude of the 1.6% increase in average travel time was considered minor.
Relatively long spacing between intersections, platoon dispersion over long distances, and very long cycle lengths were judged to be some of the reasons for this increase. The information contained in this report will be of assistance to public agencies considering the installation on emergency signal preemption systems, and to ITS engineers.
• Oklahoma Department of Transportation (ODOT) -ODOT was planning extensive improvements to I-40 through Oklahoma City when FHWA helped the state use CORSIM to evaluate two design alternatives. Comparison of the operational results between the alternatives helped ODOT identify and recommend a preferred freeway design. In addition, the animation results for both alternatives were displayed at the public meetings. The movie animation helped ODOT discuss the traffic operations and answer questions from the public.
• Orlando, Florida (FL) -Preliminary functional capabilities, which are now part of TSIS, were used to perform a corridor operational analysis on several miles of a proposed expressway/I-4 systems interchange design to be located in downtown Orlando, FL. Based on the analysis, a recommended geometric design and traffic control enhancement produced a final design that dramatically improved overall system traffic performance. In addition, a final interchange design was recommended that allowed better traffic flow during construction with an overall design and construction savings of over $10 million.  Rt. 10,. I-95 NB through the study area has typically four-lanes . As can be seen in Figure 1-1, Rt. 10 NB is situated as a loop highway branching off from I-95 NB as it comes through Cranston, RI and reconnecting to I-95 NB in Providence. Also displayed in the figure is Thurber's A enue Curve, which is a segment of I-95 NB that experiences frequent traffic incidents.
By visual inspection of the study area in Figure 1- The origin and termination points of the I-95 NB simulated corridor were chosen for two reasons. First, the beginning point had to be upstream form the I-95 NB exit to Rt. 10 NB and the termination had to be downstream from Rt. 10 re-entry point to I-95 NB. Second, the RJDOT had documented volume data for both of the exterior sections. The specific beginning and ending points of the simulated 1-95 NB corridor studied correspond to the location of the documented traffic data that was available. These points for which traffic volume exists are located approximately one mile south of the exit to Rt. 10 NB and Y2 mile north of the entry ramp from Rt. 10 NB. A more detailed discussion of the traffic volume data used for these points and throughout the network are presented in the preceding sections.

Defining the Necessary Modeling Assumptions
After a literature review was performed but before the network was modeled, a detailed list of modeling assumptions had to be formed. These assumptions must be made to limit the number of variables examined, decrease modeling time, and because certain parameters in the model cannot be modified without changing CORSIM programming language. Modifications to the encoded standard logic of the software is beyond the scope of this research. The major modeling assumptions made were: 6. That, due to a modeling protocol in CORSIM, modeling the Rt. 10 exit ramp and over pass in a way that is not exactly how this system exists will not create a difference in traffic flow characteristics through this section; 7. That all arterials directly connected to freeway ramps will not be affected by the diversion and are operating under normal flow conditions; and 8. That vehicles on the mainline will seek no other alternate route at the time of the incident other than Rt. 10 NB. The reasons for choosing these MOE's as the network evaluators was because they are typically used in traffic analysis and can be readily associated to other network characteristics, like level of service (May 1990 This section describes the information available and the methods of extrapolation used to generate the complete modeling data set. The section is divided into two sub-sections. The first deals with the roadway geometry and how it was obtained and modeled. The second covers the roadway and ramp traffic volumes and how they were obtained, extrapolated, and modeled.

Roadway and Ramp Geometry
The network geometry and lane configurations were obtained from satellite aerial photos. These aerial photos were downloaded from the Microsoft™ Terraserver™ website (www. terra crvcr.corn). Aerial photos were chosen because they were readily available and because they provided the high level of detail needed to accurately model the network, especially with regards to lane configurations. The photos were downloaded at the scale of one-inch equals 300 feet, printed out, and then assembled on a 6x8-foot piece of foam board. The photos were then overlaid with 45 transparency paper. The transparency overlay made it was possible to mark the map while having the option to remove the mark if modifications were necessary. This map enabled researchers to obtain the following network characteristics: • mainline and alternate route orientation and geometry, • number oflanes on all segments of both routes, • exit and off ramp geometry, • merge, diverge, and weaving area configurations, and • acceleration and deceleration lane lengths.
With the map and transparency overlay assembled, a custom coordinate system was established so that the roadways and ramps of the network could be specified as nodes and links. The custom map was mounted on a wall and labeled with the coordinate system as it was developed. The coordinate system had to be measured in linear feet because that what is required by CORSIM. The positive horizontal, or x, direction was specified from west to east beginning approximately 50 feet to the west of the western-most point on Rt. 10. The positive vertical, or y, direction was specified from south to north beginning approximately 100 feet south of the southernmost point on I-95. This mapping system allowed researchers to view all the network data that was being recorded and readily locate and modify any discrepancies efficiently.
After the number coordinate system specified, the network had to be defined as a series of nodes and links. "Nodes" and "Links" are terms used to describe the features of CORSIM that are used to model simulated networks. They are the most important design characteristics of the roadway network because all traffic volumes, 46 roadway characteristics (number of lanes, free flow speeds, geometry), turning movements, and incidents are specified at the node or on the link. A node is placed at the following locations in the network: (1) wherever the network geometry changes from straight to having curvature; (2) at the boundaries of the network where traffic either enters or exits the system; (3) at all exit or entry ramps ; and ( 4) wherever lanes are added or dropped. Links are basically the simulated roadway that connects the nodes. The entire study area roadway network was defined on the network map in the established coordinate system as nodes and links. Before the actual numbers were assigned to each node the researchers became familiar with the protocol for node labeling in the CORSIM User's Manual. The nodes for this modeling procedure are characterized as follows: • External Nodes -an outer boundary, a node from which traffic enters or exits the system, represented in ITRAF by a hexagon and numbered 8000-8999.
• Internal Nodes -one completely embedded in the model. An internal node is represented by a circle and numbered from 1-7 50 It was determined that external nodes would be needed for every extreme entry and exit point of the network. Specifically, an external node would be needed at the beginning and ending of the two routes (I-95 and Rt. 10) and at the beginning of all entry ramps and at the termination of all exit ramps. Also it was decided that it would be advantageous to create a sequential numbering system for all the nodes. This numbering system was utilized because the output files arrange the output data by link, and the links are listed sequentially by order of beginning node. By instituting a 47 sequential nodal representation from the southern-most points on each roadway a sequential output file is also created. A sequential output file was deemed valuable for output data collection and model validation phases. By increasing the node numbers by five, researchers are able to easily add nodes in places either unforeseen or desired later without eliminating the numerically ascending sequence established. Tables C-1 and C-2 located in Appendix C list the coordinates for all the nodes as scaled off the study map and estimates the length of the links created between the nodes.  At the beginning of the network modeling, it was thought that Rt. 10 NB and 1-95 NB could be simulated as they exist in the real world. However, one uninterrupted flow facility cannot extend from the same type of facility. FRESIM does not allow an uninterrupted flow freeway (Rt. 10) to extend from another uninterrupted flow freeway (I-95). One of the major challenges of developing the simulation model arose while entering the network geometry.
The first attempt to overcome this obstacle was to utilize interface nodes.
Interface nodes are used to seamlessly link the two models within CORSIM, FRESIM and NETSIM. Researchers experimented with deceiving the program by placing an interface node between the two "freeways". By modeling the network this way it was thought that the model would essentially resolve that the system being modeled went from freeway (I-95) to urban street network (Off Ramp to Rt. 10) and then back to freeway (Rt. 10). By using this short NETSIM link and interface nodes it was believed that the software would be manipulated into accurately modeling the network, as it existed. However, after initial tests were perfom1ed on the network established as such, it was determined that the simulated network was not performing as intended. This was concluded because the output statistics were determined to be Un-reasonable and because unusual vehicle behavior was noted while monitoring the vehicles traveling across the network using the TRAFVU. It was determined that the network would have to modeled as two separate highways as opposed to one system as it exists in the real world. 53 Modeling the network as two separate roadways was determined feasible because the network only had to be "broken" at three points. The three points are where Rt. 10 branches off of I-95 NB, where Rt. 10 enters I-95 NB at the southern most portion of the study area, and at the point where Rt. 10 returns to I-95 northern most portion of the study area. It was now necessary to calculate exactly how many vehicles would be exiting I-95 NB to Rt. 10 NB and likewise back to I-95 NB at the termination Rt. 10 NB for any given traffic volumes modeled. The procedure for determining the exiting and entering volumes for a given simulation situation is described in detail in the next section. In order to accomplish the separation of the network with in the model, the network was severed at its connecting ramp links. At the break between in the ramps external nodes were placed. These additional external nodes required traffic volumes based on the level of traffic volume modeled. Further explanation is provided in the validation sections found later in this chapter.
Separating the two roadways was not found to detract from the integrity of the model.

Roadway and Ramp Traffic Volumes
With the network geometry established, the next step was to determine the traffic volumes to be entered at each entry node. The RIDOT was contacted in order to ascertain what traffic information could be provided for the two main roadways in the study area and the exit and entry ramps that connect to each. RIDOT provided their data submission sheets for the 1998 FHW A Highway Perfomrnnce Monitoring System (HPMS) Report. 54 The HPMS data was reviewed to determine which data was pertinent to this research. At first inspection, the HPMS data appears awkward because of the way that it is organized. Each section of all public roadways has its own page of data, but the report is not organized by roadway titles (For example: all of I-95 data is not grouped together, but rather, data for this roadway is dispersed among the other major highways in the state). The page for each section of roadway includes many statistics relating to the traffic on the section. However, for this research, only 3 section statistics were needed from the pages; average annual daily traffic (AADT) , the Kfactor (K), and the directional distribution factor (D). With these three factors the directional design hourly volume (DDHV) was computed and used as traffic volume input data for the model.
The DDHV typically represents the thirtieth highest peak hour volume of the year, and is calculated using the following equation: Where: proportion of daily traffic occurring during the peak hour, expressed as a decimal proportion of peak-hour traffic traveling in the peak direction, expressed as a decimal Once the peak AM hourly volumes were determined multiplying these volumes by a factor relating to the new volume desired could attain any variation on this volume.
For example, if the half peak volumes were needed then the peak volumes would be 55 multiplied by a factor of one-half.  The HPMS data attained was sufficient enough to perform an extrapolation that was able to generate a complete traffic volume input data set. It was imperative that every external entry node had a representative DDHV. Extrapolating the HPMS data began by developing a traffic volume map of the network. This map is composed of the two mainline and alternate routes and all the exit and entry points in the network. The locations in between all entry and exit ramps in the network were labeled as critical points so that they could be linked to the calculations formed in the spreadsheet. These critical points correspond to every place along each route where the traffic volume will change due to exiting or entering traffic volume. For example, if there is an exit ramp, the traffic volume before is a critical point, and the volume after is a critical point because the volume will change based on some percentage of the vehicles exiting. The calculations used were based on deductive reasoning. For example, if a traffic volume upstream of an entry ramp was known and the volume directly downstream from the same ramp was known, then the difference in the volumes was determined equal to the amount of vehicles that should enter at that ramp. Using the traffic volume map displayed in Figure 3-5 a spreadsheet was created so that ranges of volumes and percentages of vehicles exiting could be manipulated until a representative data set was found. Since only limited documented data was known for each roadway and no documented data was available for the majority of their entry and exit ramps it was necessary to establish a method through which the volumes could be easily modified. Table 3  Rl-1 0-1 RJ-10-2 RJ-1 0-3 Rl-10-4 Rl-1 0-5 Rl-10-6 Rl-10-7 Rl-10-8 RJ-10-9 Rl-10-10 Rl-10- Because of the modeling challenge described earlier, it was necessary to determine the traffic volumes at the three exit points that correspond to the points where the network was divided. Specifically, it was necessary to calculate how many vehicles exit to I-95 NB from Rt. 10 NB at the southern portion of the study area, to Rt. 10 NB from I-95 NB, and to I-95 NB from RT-10 NB to I-95 NB at the termination of Rt. 10 NB at the northern portion of the study area. These specific volumes were needed to link the two roadways. With these volumes, the network could be modeled as close to real-world conditions as possible. The traffic volumes at these three points could also be calculated for any given route diversion situation.  With the AM peak hour volumes determined, the geometric network of nodes and links assembled in ITRAF was revisited in order to add the entry volumes at the necessary entry nodes. Since the geometry of the network had been established earlier, the traffic volume data entry process was straightforward. To enter traffic volumes in the ITRAF GUI user environment the user must select the link between an exterior node and the first internal node of the roadway link so that a data entry window appears. In this window the volume of tra591c to enter the system for a given time period is specified. At this time, the user can also specify the percentage of trucks for the roadway's segment and any predetermined lane distribution. If no lane distribution is specified, CORSIM randomly assigns the vehicles to their initial lanes.
Lane distributions were not assigned because this research seeks to utilize the stochastic nature of the model. Time periods are user specified through the simulation time interval and will be explained in greater detail in the next section. Figure 3-6 displays the data entry process utilizing ITRAF.

3.t.5.3 Simulation Time Considerations
There are three main considerations when specifying time constraints within CORSIM. First is the pre-simulation time; the second is the overall simulation time period; and the third is the number of time periods within the overall simulation time.
The pre-simulation time is the time prior to the actual simulation. During this time, the network is attempting to reach equilibrium, that is the number of vehicles entering the equals the number of vehicles exiting. During this period no network statistics are generated, but an indication is given at the conclusion of the allotted time. At the end of the specified time, CORSIM infonns the user whether or not the network has reached equilibrium. This time can be left open allowing CORSIM to run until the network reaches equilibrium, at which point it will automatically start into the actual simulation period. For all simulations ruri) during this research, the pre-simulation time was not specified so that no simulation was begun before the network had reached equilibrium.
The actual simulation time interval was originally set at one-hour. This time was chosen because it was the standard found during the literature review. However, after a preliminary incident effect investigation was perfom1ed, it was decided that the one-hour time period was not allowing the traffic to fu lly recover after the event. Due to the degree of lane blockage and the duration of the incidents to be modeled, it was determined that a two-hour simulation time would be required.
The time intervals within the overall simulation time period are used for two purposes. First, by specifying time intervals within the total simulation time the user 67 creates points at which CORSIM will generate cumulative network statistics. The two-hour simulation time used in this research was comprised of twelve ten-minute time periods. This means that for every ten-minutes of simulation time a cumulative table of network statistics were generated. These ten-minute reports aided researchers in assessing how the model was performing throughout the overall simulation time, especially as incidents were introduced. The specification of time intervals also enabled the researchers to insert various changes to the model during the course of the simulation. Specifically, researchers could vary volumes, percent of traffic exiting, and lane blockages for any number of time intervals created.

Simulation Random Seed Numbers and Replications
As discussed in the introduction, one of the assets of CORSIM is that it is a stochastic model that generates traffic volume patterns based on random seed numbers. The significance of the stochastic nature of the model is that the same traffic volumes are able to travel through the network in different patterns and produce varied network statistics. For every situation modeled, three simulations were performed, each with a different random seed number. This produced different outputs for the same traffic volumes and allowed for statistical analysis on the performance measures.
The random seed numbers used were one (1), thirty-three (33), ninety-nine (99). conceptual validation and operational validation. Conceptual validation is a process of assessing the theoretical and software models against sound and accepted theoretical foundations. The operational validation process consists of comparisons between model operational predictions and measured real-world system operational behavior (Benekohal 1991 ).
This section explains the observational and operational validation of the specific simulation model developed for this research . An operational validation criterion includes threshold values on the quantitative measures of consistency between model results and real-world data. A model is never an absolutely accurate translation of the real-world system. Therefore, criteria for validation must be less than one hundred percent correspondence to the real-world system. However, criteria 69 must be established so that a level of certainty in the model can be stated.

Observational Observation Utilizing TRAFVU
The first step in the validation process was to run the simulations and then view them with the movie player companion software, TRAFVU. The validation simulations were run under the normal AM peak traffic volume conditions.
Simulations run under nom1al conditions (experiencing no incidents) will be referred to as base line simulations throughout this section. The base line simulations will also employ the random seed numbers specified in the previous section (1 , 33 , and 99). By using this movie player companion software, researchers were able to assess how accurately the model was representing the actual conditions.
The first problems that came to the attention of the researchers during the observation of the base line simulations were the human en-ors in the modeling of the roadway geometry and lane alignments. If the lanes were not aligned correctly while modeling the network in ITRAF, then the traffic would become congested at the point of misalignment as if there was a sudden bottleneck. TRAFVU was excellent in exposing mistakes that may not have been evident by only examining output data.
Once these geometric set backs were resolved, the base line simulations were run again and the previously problematic areas were monitored. After the visual analysis rendered no problems and the model appeared to be functioning as intended, the next validation phase was begun.

Operational Validation Utilizing the Floating-Car Technique
The first step in the operational validation of the model was to establish the If the field data sets were found to be equal to the simulation data sets, based on the statistical analysis, then the experiment would continue with the model 72 unmodified. Conversely, if the field data sets were found to be significantly different from the simulation data then the model and all assumptions would be re-evaluated.
At that point model modifications would be made, and the validation process would be repeated.
Once the decision criteria for the operational model validation were established the technique for field speed data collection was chosen. After reviewing methods described in various textbooks it was decided that the floating car technique would be adequate. The floating car technique is a practical, economic, and a widely used method in obtaining field data for roadways and validating traffic models (May 1990).
The floating car technique consists of driving a test vehicle in the traffic stream and behaving as a standard vehicle. For this validation analysis the test vehicle would traverse both roadways (Rt. 10 NB and I-95 NB) in the study area for a specified distance of one mile. The one-mile sections correspond to specific links within model.
By using closely related roadway segments researchers gained insight into how the model output is comparing with the observed field data.
In order to obtain a representative speed for the segments ofl-95 and Rt. 10 the floating car technique requires that the test vehicle behave like an "average" vehicle (May 1990). An "average" vehicle was defined for this research as a vehicle traveling in the second lane from the right, typically called the "travel lane" that attempts to pass one vehicle for every vehicle that passes it (May 1990). Both roadways were driven during the peak AM period under normal flow conditions (no incidents). The speeds noted for the three floating vehicle runs and the speeds obtained from the simulation runs can be found in Table 3-4 and Table 3-5. Table 3-4 contains the data 73 complied for 1-95 NB and Table 3-5 for Rt. 10. Also shown at the bottom of each of these tables are the results of the statistical tests discussed earlier in this section.
The simulated speeds shown for specified links in the network were obtained from the output files generated by CORSIM. The speeds obtained from the floating car analysis are for similar links on the actual roadways. Meaning the actual driven links begin about were the simulated links start and end one-mile after this start point.
The simulated network lengths are more specific because they are generated in the software based on the node-to-node distances and the radius of curvature specified.
For the floating car analysis only one-mile lengths were used because of the accuracy of the measuring equipment. The speeds for the floating car analysis were found by dividing the distance traveled by the time taken to reach the length.
74  The F-value and p-value detem1ined and stated in Table 3-4 for the operational validation of the I-95 NB roadway indicate the following: • that the data is comparable because the variances are not significantly different, and • that the means of the data sets are not significantly different.
The F-value and p-value determined and stated in Table 3-5 for the operational validation of the Rt. 10 NB roadway indicate the following: • that the data set is not comparable because the variances were found to be significantly different, but that, • based on the t-test performed, the means of the data sets are not significantly different.
This set of statements show that the model passed all aspects of the operational validation except the test on the comparability of the data sets for the Rt. 10. A possible remedy for this is to obtain more samples from the field to be incorporated into the F-test and t-test. However, due to time constraints this was not feasible. The model was considered to have passed the operational validation phase and was ready to be modified according to the incident/diversion experiments designed.

Comparison with the Results of Highway Capacity Software Analysis
The HCS software package has become the standard software for replicating These four sections were chosen because they are important in terms of how the network will function with regards to the route diversion analyzed.
The comparison proceeded by detennining the LOS for the four sections using RCS and CORSIM. The LOS is a letter designation , ranging from A to F, that describes a range of operating conditions on a particular type of facility. The LOS 78 was provided directly from HCS when the prevailing traffic and roadway conditions were entered. The CORSIM output data obtained from the output files for the link densities was converted to LOS designations by utilizing the tables for in the 1997 HCM (TRB 1997). It is important to note that the volumes and heavy vehicle percentages entered into HCS were identical to those entered into the CORSIM model and that the default values found in HCS were not adjusted.
After the LOS for each of the four sections was determined, the results were compiled in Table 3-6. Table 3-6 displays that the two techniques produced identical LOS designations for the I-95 NB sections (la and 2). However, the Rt. 10 NB sections show a discrepancy between the two designations. These results for the LOS are comparable. A platform for future research could be to determine why the discrepancies are occurring for the Rt. 10 sections analyzed.

3.t.6.4 Summary of Model Validation Process
In summary, the validation process involved three phases. The first was the observational phase, which entailed viewing the model in TRAFVU. The second was the operational phase, which consisted of a comprehensive statistical analysis. The third was a conceptual comparison between the standard freeway computational practice and the model used in this research. Although the validation was time consuming and difficult, it was essential to the modeling process. If the model was not validated, then it would be impossible to identify if it was accurately portraying the actual conditions it seeks to describe. Once the validation process was complete, the analysis methodology was addressed.

3.2.l Introduction
This section will continue the description of the research methodology initiated in section 3.1. This section is divided into two major subsections that correspond to the two phases of analysis utilized during the research. The first phase presented explains the characteristics, or factors , associated with a traffic incident, how each of these factors was modeled in CORSIM, the preliminary experimental design and analysis that determined which of these factors would be fixed or varied in the second phase. The second phase involved expanding the preliminary experimental design to include greater ranges of variability for the factors found to be significant in the phase one analysis. This two-phase analysis technique was employed because it empowered researchers to reduce the range of variability for less significant factors while increasing the variability of those factors found to be most significant. By reducing the variability of certain factors the final experimental design was reduced, and modeling and simulation-processing time was saved.

Route Diversion
Defining the factors for any experiment is a tactful step because a balance must be attained between the inclusion of all possible factors that could affect the outcome of experiment in some infinitesimal way and the failure to incorporate enough factors to accurately represent a given situation being tested. At the onset of this study it was 82 noted that the factors chosen must be comprehensive enough to accurately represent the incident/diversion events tested, but, at the same time, be succinct enough to ensure that they could be quickly identified by a TMC at the time of an actual incident.
As a guideline it was stated that: " the factors should be able to be quickly assessed through visual inspection over a video surveillance system comparable to the system employed by RIDOT". This guideline was established because incident management and traffic diversion focus heavily on decision and deployment time. Excess factors being evaluated at the time of an incident cost a TMC operator excess minutes.
Additional minutes in assessment and response can decrease the safety of those involved in the incident and extend the related traffic congestion by hours. It was also determined that the factors will be analyzed as independent of each other.
Traffic incidents range in severity from one-vehicle breakdowns on roadway shoulders that last Jess than 10 minutes to events involving multiple vehicles that eliminate or severely reduce a roadway segment's capacity for hours. The scope of this research encompasses the investigation of the effects of "minor" traffic incidents only. At this point, it was important to establish the definition of a "minor" traffic incident as applied throughout this research. For this research, a minor traffic incident is any incident that blocks three or Jess Janes of traffic on a segment of freeway (I-95 NB) for a time less than or equal to forty minutes. This definition was fommlated The next factor determined was the traffic condition, or level of traffic volume the network is operating at the time of an incident. This factor would significantly impact any decision that a TMC operator may make with regards to deploying a traffic diversion strategy. The notion that traffic volume would make a considerable impact on the network MOE's could be assumed intuitively, but the specific effect of distinct levels of traffic volume must be assessed. With the effects of various levels of traffic volume evaluated, TMC operators can confidently deploy or refrain from deploying a diversion strategy based on sound evidence. Also, traffic volume is a factor that can be readily assessed by a TMC operator. A TMC could make use of roadway detection devices like loop detectors or video detection software to immediately estimate a level of volume for the network.
The last and most difficult factor to characterize is the level of traffic diversion. For this research, this factor was defined as an exact percentage of traffic exiting from the mainline to the alternate route. The reason that this factor is challenging to define is because it would be impossible to divert a specific percentage of traffic to the alternate route during an actual incident situation. However, the percentage of traffic to be diverted could be linked to the strength of message disseminated at the time of the diversion.
With the four factors determined the focus was shifted towards the limits of variability for each factor. The range for the percentage of traffic to be diverted was set from zero to 20 percent above the nom1al percentage to exit at that point (10 percent). This range was instituted for two reasons. The first reason is that the single Jane exit ramp to Rt. 10 NB at peak AM volumes could not accommodate more traffic than 30 percent (20 percent for the diversion and 10 for traffic already exiting) of the I-95 NB through traffic at peak AM volumes. The traffic at this time would exceed the ramp's capacity and cause a congested situation. The second reason is that it has been noted in earlier research that CORSIM has difficulties simulating traffic exiting a freeway under severely congested conditions at percentages above 40% (Cragg and Demensky 1994).
The traffic volume range was set from AM peak to Yi AM peak conditions.
These limits were chosen for the volume because they could be easily calculated based on the earlier AM peak volume calculations and because they represent very distinct conditions that the roadways and ramps experience daily. The volumes utilized for this phase of experiments and the volumes used for the next phase are referenced in Appendix D. All volumes are based on the traffic flow map established earlier in The ranges for the level of incident and duration of incident are spelled out in the definition stated for minor traffic incidents. The ranges for the duration and level 85 of incident were set at 20 to 40 minutes and zero to 20 percent respectively. Table 3-7 provides a summary of the factors to be modeled and the limits of their variability. 86

J.2.2.2 Traffic Incidents and Diversion in CORSIM Software
Once the roadway network was established, entering a traffic incident situation in the ITRAF GUI was straightforward. The data required to specify the occurrence of a traffic incident include the following: • the link on which the incident occurs • the location of the incident on the link • the length of roadway effected by the incident • the time the incident begins and ends • the rubberneck factor The actual freeway in which segment the incident will occur is the "Thurber' s Avenue" curve segment of I-95 NB. In the simulated network, this segment is the curved link between nodes 15 and 20. The incident was positioned at northern most end of this simulated link because on the actual segment this is most poorly designed section of the segment.
The length of the simulated roadway link that is affected by the incident is related to how many cars are involved in each lane of the traffic incident. The CORSIM manual advises that the incident be modeled as if each car is 20 feet. For this research a lane blockage consists of two cars. In addition to this length an additional 20 feet ahead of the incident should be included as a "rubbernecked zone".
The rubbernecked zone was created by specifying an incident in the lane(s) to be blocked that consists of only a rubberneck factor (no physical blockage). The rubberneck factor is a characteristic associated with driver' s tendency to slow down to observe traffic incidents. The factor was fixed at 10% because prior research 88 determined that this setting most accurately describes what has been recorded for incident data (Cragg and Demensky 1994).
The total simulation time was two hours with the incident occurnng ten minutes after the start of the simulation. This incident start time was chosen because it provided a period of normal traffic conditions to compare with subsequent periods of abnormal traffic conditions. As stated earlier, the limits incident duration range from 20 to 40 minutes. This meant that any incident modeled would be concluded before one hour of simulation has elapsed. By concluding the incident within one hour of simulation time the network would always have at least one hour to attempt to recover from the incident situation.

Overview of the 2k Factorial Statistical Analysis Technique
The 2k factorial experimental design was chosen for the preliminary analysis because it allows all factors for a given experiment to be analyzed at "high" and "low"

Application of the 2k Factorial Statistical Analysis Technique
The first step in the application of the 2k Factorial Statistical Analysis Technique was to determine the two distinct levels at which each factor would be varied. It was decided that the factors be set at their high and low limits in accordance with the procedure for the analysis. MINIT AB™ requires that numerical labels be attached to the levels of the factors. For the ranges specified in this phase, the number -I was attached to the lower limits and the number + 1 to the upper limits. Table 3-8 displays the levels at which the factors were fixed and how they were coded for their entry in MINIT AB™.
90 For these preliminary experiments it was necessary to determine the volumes at all the critical points in the network for both the peak and Yi peak AM traffic and for both the 0% traffic diversion and 20% traffic diversion scenarios. This was necessary because the input volumes in each of the situations are unique and will significantly affect both the mainline and the alternate route. Appendix D contains tables that list, in accordance to all the critical points and entry and points established in Figure  Each combination of the factors at two different levels specified represents one simulated situation. For example, the network operating at AM peak traffic conditions, experiencing a three-lane incident lasting 20 minutes with no diversion deployed was one situation. The total number of situations that can be formed from four factors at two levels is 16. Each of these 16 different situations was modeled and replicated by assigning three distinct random seed numbers. After replications were created the total number of simulation runs for Phase-I was 48. The speed data needed was obtained from the network statistics found in the CORSIM output files and recorded in spreadsheets. These spreadsheets are located in Appendix E, F, G, and H.

92
These four appendices contain all output data collected for both phases of experiments and for all three MOE's. The output data for the Phase-I experiments is found at the beginning of Appendix E, the appendix containing the output data for the AM Peak situations modeled, and Appendix H, the appendix containing the output data for the Yi AM Peak situations modeled. Actual CORSIM output files could not be printed out and included because each simulation output file consists of more than 100 pages.
The first step in the analysis was to collect and organize the network speed output data from each of the 48 CORSIM simulation output files. Since the network was modeled as two roadways, the network speed had to be calculated as a weighted average. A cumulative network average speed was found by dividing the sum of all the miles traveled on the two roadways and their ramps by the sum of the time taken to travel these respective distances.    The calculated p-values in Table 3-11 show that the volume, degree of incident, and duration of incident had significant effects on the response. However, the level of diversion did have a marginal effect on the response. Also found to be significant were the following combinations of factors: volume and incident, incident and duration, and volume, incident, and duration.

Recommendations for Phase-II
Based on the preliminary experiments performed and the subsequent analysis it was decided that only three of the four factors should be varied to greater degrees between the established limits. However, since the diversion was found to have had a marginal effect on the MOE it was decided that it should be fixed at its stated limits for further evaluation. These decisions are based on the ANOV A analysis performed and the previously stated level of significance. The expanded experimental design executed in Phase-II includes the factors and levels described in Table 3-12.

Application of the Expanded Experimental Design
The Phase-I analysis showed that three of the four individual factors had significant impacts on the cumulative average network speed. The ANOV A table presented in Table 3-11 led to the decision to vary the factors as described in Table 3 The total network delay was found using the following equation: Total Network Delay (hours) = (Rt. 10 Total Network Delay+

1-95 Total Network Delay )( I hour I 60 minutes) (Equation 3-3)
With these equations in place the next step was to run all the simulations and acquire the three network statistics for each roadway for every simulation.

Definition of Decision Criteria
In an experimental procedure it is essential to state decision criteria before the minutes with traffic diversion. The decision criteria were put in place to objectively select one of two options. The options are either: (1) the diversion was beneficial to the overall system and is warranted for the given situation, or (2) the diversion was not beneficial to the overall system and is not warranted for the given situation. This t-test provided initial information on the effect the diversion had on the MOE's. Based on the three replications for each situation, the t-test calculates a test statistic that incorporates the variance and mean of each data set. By taking into account the mean and the variance, the t-test provides a measure of comparison that is more representative than merely comparing the sample means.
For the diversion to be considered wan-anted, the Ho stated in the t-test must be rejected and the difference in the means must favor the situation with diversion. The difference in the sample means was evaluated using the percentage differences and the actual differences between means of comparable data sets. Specifically, diversion was considered warranted if: the cumulative average network speed had increased and total network delay had decreased with diversion. On the other hand, diversion was considered not wan-anted if the cumulative average network speed had decreased and total network delay increased with diversion.
Also formulated during this part of the analysis were three General Linear Model ANOV A tables. The three tables are based on the three different responses, or MOE's, recorded for each of the replications modeled. These ANOVA tables display the p-values for each factor and the various interactions between the factors. The hypothesis and rejection criteria for these three tables are identical to those indicated in for the ANOVA analysis utilized in the Phase-I analysis of this research.
Essentially, a p-value less than 0.05 indicates the factor, or interaction of factors , is significant.

Presentation of Results of Phase-II Experiments
The results of the Phase-II experiments are summarized in Tables 3-13 through 3-22.           respectively. Each set of tables in these four appendices was arranged by incident event and MOE being reported. These appendices are arranged in the following order: first the calculations for the cumulative network average speed are presented for every situation modeled, then the calculations for the cumulative network average travel time are presented for every situation modeled, and, finally, calculations for the total network delay are presented for every situation modeled. The data presented in the following tables will be discussed in the next chapter.
With this data compiled and analyzed the next step was to discuss the significance results and determine when the diversion strategy is warranted. This task is addressed in the next chapter.
116 CHAPTER 4. DISCUSSION OF RESULTS

Evaluation Based on the Decision Criteria Defined
The discussion of the results centers on the previously stated decision criteria.
The decision criteria put in place in the previous chapter was used to determine if the traffic diversion modeled had beneficial effects on the network. In order for the diversion strategy to be implemented for a given traffic condition and specific incident situation the following decision criteria was established: I. The p-value determined for the Paired t-test performed must indicate that there is a significant difference between the means of two comparable MOE data sets. A p-value of less than or equal to the stated level of significance (0.05) indicates a significant difference between the means of the two data sets.
2. The percentage difference between the means of the MOE data sets without the diversion strategy deployed and the data sets with the diversion deployed meet the following criteria: positive for speed (i.e., the cumulative average speed increased for the situations with the diversion) and negative for delay (i.e., the total delay decreased for the situations with diversion) If these two criteria were met then the diversion was considered beneficial for the entire network. If either criterion was not met, it was concluded that the diversion had no impact or a negative impact on the network.

117
The subsequent sections of this chapter discuss the results presented in Tables   3-13 through 3-22. Tables 3-13 through 3-20 list the results for the paired t-tests   performed on each set of comparable data. Tables 3-21 Tables 3-13 through 3-20 are presented along with a summary table of the recommended diversion strategy for every situation tested is referenced in Table 4-1.

Conditions
The results for the AM peak traffic conditions, as displayed in Tables 3-13 Table 3-13 by a p-value of 0.018 coupled with the percent difference in the means of -38%. The p-value determined for this situation is below the level of significance and the percent differences for the network MOE ' s are demonstrating that the network operated better without diversion.

I l9
Another characteristic of the results for the AM peak traffic conditions is that the standard deviations calculated for the MOE's for the situations with diversion are, on average, greater than those for the simulations without the diversion. For example, the average standard deviation for the speed data from the AM peak data listed in Table 3-13 with diversion is 7.38 mph verses 1.88 mph without diversion. This is noteworthy because it conveys that the simulations are relating the random nature associated with traffic flow. An actual traffic diversion could be expected to yield random impacts on a network because every element involved, human controlled vehicles, is random. Future research could utilize more repetitions of each experiment to ensure that the large variances found are not due to the extreme cases in the distribution.

Conditions
The results for the simulations modeled with % AM Peak traffic conditions, as listed in Tab les 3-15 and 3-16, reveal different effects on the MOE's than those 123 observed for the simulations modeled with AM peak traffic conditions. For every MOE the situation that consisted of one lane being blocked for 20 minutes reveals that the tests indicate that the diversion was beneficial to the network. However, for most of the other cases the p-values are greater than the level of significance of 0.05. This indicates that there was no significant difference in the means of the comparable data sets. The p-values found for the speed data, as listed in Table 3 Table 3-15 , the percent differences corresponding to each of the above p-values are: 1%, 10%, -3 %, -1 %, 9%, -4%, 9%, 13%, and 34%. In this case the p-values and the small percent differences clearly show that the diversion had no significant impact on the network evaluators. One reason for this phenomenon could be that there more replications are needed to provide more representative data sets for each situation.  It is presented to further illustrate the relationship between the degree of incident (number of lanes blocked and duration of blockage), level of diversion (0% or 20% diverted to the alternate route at the time of incident) and the cumulative average network speed for each scenario. This figure reinforces that, at % peak traffic conditions, the diversion had slight positive or no significant impacts on the network evaluators. The figure demonstrates this by displaying the relationships between the comparable data sets. Since the diversion had slight positive effects on the network evaluators, the curves for the diversion are slightly higher for each incident event than the curve for the situations that did not utilize diversion. This figure, coupled with the discussion throughout this section, provides evidence that the diversion strategy could have positive effects on the network at% peak traffic conditions.

Conditions
The results for the simulation modeled with 2/3 AM Peak traffic conditions demonstrated that the diversion had primarily positive effects on the network MOE 's.
The evidence for this statement is found in Tables 3-17 and 3-18 by examining the pvalues and the percent differences found for four of the nine cases indicate that there is a positive significant difference between the comparable data. The effect of the diversion is highlighted by large, positive, percent differences between the comparable means for the network speeds and the delay times. For example, as seen in Table 3-18, the cases that indicated significant differences for the delay times indicated percent differences of -7%, -61 %, -78%, and -85 % (negative percent differences are indicating that the delay with the traffic diversion is less than the delay without). There were four cases for which the diversion had significant positive effects on the network MOE's. These cases were: when the level of blockage was one lane and the duration of incident was 20 minutes, when the level of blockage was two lanes and the duration of incident was 20 minutes, when the level of blockage was two lanes and the duration of incident was 30 minutes, and when the level of blockage was two lanes and the duration of incident was 40 minutes. Furthermore, two instances were indicating that the diversion had marginal positive impacts on the network. These two situations were when the degree of lane blockage was three-lanes and the duration of the incident was 30 and 40 minutes. The evidence for this statement is found in the percent differences found for the comparable data sets. For the speed data the percent differences were 24% and 25% and for the delay data the percent differences were -11 and-17% The standard deviations are less, on average, than those calculated for the simulations modeled with AM peak and % AM peak traffic conditions. For example, the average standard deviations for the delay for the AM peak and 3 14 AM peak for the cases with diversion are 597.14 and 449.08 hours, respectively, while the average standard deviation for the delay for the 2/3 AM peak for the cases with diversion is 79.00 hours. This is indicating that the lane-changing, weaving, and merging problems present at the higher volumes are significantly Jess at this volume level.
The reason for this could be that the traffic being diverted is below the threshold volume that causes the merging, diverging and lane-changing difficulties observed at the higher traffic volumes tested. This indication suggests that 2/3 AM peak could be a threshold value for which greater volumes of traffic should not be diverted.
In addition, even if the paired t-test's indicated no significant difference, the percent differences in the means are indicating beneficial impacts for all cases. A typical example of this can be seen in the situation modeled with 2 lanes blacked for 30 minutes. For the speed data for this situation, as seen in Table 3-17, the t-test found was 0.074 indicating no significant difference in the data sets. However, the percent difference of 57% indicated that the diversion did have a beneficial effect on the network. The data from the simulation model indicates that at 2/3 AM peak traffic conditions the diversion strategy tested had the most beneficial impacts on the network for the set of traffic conditions modeled.
128 Figure 4-3 is presented to further highlight the relationship between the degree of incident (number oflanes blocked and duration of blockage), level of diversion (0% or 20% diverted to the alternate route at the time of incident) and the cumulative average network speed for each scenario. Figure 4-3 was constructed using the network speed data found in Table 3

Conditions
The results for the simulations modeled with Yz AM peak traffic conditions, as seen in Tables 3-19 and 3-20 do not continue the trend that the 2/3 AM peak data indicated. That is, in most cases the diversion had no significant impact on the network MOE's. However, in two cases, the p-values and percent differences found indicated that the diversion did have a significant beneficial impact on the MOE of delay. The two cases were: the situation that consisted of a degree of incident of onelane blockage lasting for 40 minutes and the situation that consisted of a degree of incident of two-lane blockage lasting for 40 minutes. The respective p-values for these two cases were 0.048 and 0.050 and can be seen in Table 3-19. The fact that the diversion did not have any considerable negative effects on the MOE's could be indicating that traffic conditions equal to, or less than, Yz AM Peak could manage a greater percentage of diversion. The evidence for this is that the average percent difference in the means for the speed data for the 9 cases, as seen in Table 3-19, isl.6%. • when operating at AM peak traffic conditions, the effects of the diversion on the MOE's were significantly negative; • when operating at % AM peak traffic conditions, the effects of the diversion on the MOE's were slightly positive or slightly negative; • when operating at 2/3 AM peak traffic conditions, the effects of the diversion on the MOE's were significantly positive; • when operating at Yi AM peak traffic conditions the effects of the diversion on the MOE's were marginally beneficial to insignificant.   Based on the results of this research, a possible configuration for these types of protocols was developed. This set of instructions is a significant product of this research. The protocol developed is referenced in Table 5-1.
The incident management protocol presented in Table 5-1 was developed directly from the evaluation of the results of the simulation experiments performed.
The decision criteria stated enabled researchers to state, based on quantitative MOE's, if the traffic diversion strategy tested is warranted for 27 different minor traffic incident situations. The protocol displayed in Table 5-1 presents a set of procedures that could be followed by an A TMC operator at the time of an incident. The procedure developed includes: numbered sequential tasks, descriptions of the action to be taken to complete each task presented, and a recommended completion time for each task. The procedure also indicates the action that should be taken if the diversion is determined not warranted or if the incident situation does not resemble one of the 27 situations analyzed.    is not warranted, then the operator shou ld continue to monitor the incident and provide the emergency team with ACTIVATE whatever assistance is needed. Also, the VMS and HAR 12 Diversion should continue to relay the initial message of incident 10 description and caution . These messages should remain Strategy until the comp lete clearance of the incident. If the incident is under control and being managed the operator should note any vehicle diversion to Rt. 10 without the prompting of the TMC. Based on the situation described by the 3 characteristics and the reference tables it is determined that the diversion is warranted, and then the operator should activate the pre-assigned messages for the HAR and VMS in the case ACTIVATE that a diversion is warranted. This language should be strong enough to divert 20% of the I-95 NB traffic to Rt.

13
Diversion 10. These messages should remain until the complete 30 Strategy clearance of the incident. If the incident is under control and being managed the operator should note any vehicle diversion to Rt. I 0 without the prompting of the TMC. The operator should also continue to monitor the incident and provide the emergency team with whatever assistance is needed.    The results of the analysis performed on the output data from CORSIM enabled researchers to form a route diversion strategy for the network studied. The developed diversion strategy is based on the determined impacts of the factors modeled on the network evaluators. It was found that the diversion is not warranted for the majority of the situations tested. However, a significant finding of the research was that at 2/3 AM peak traffic volume the diversion was found to having beneficial effects on the network for more of the incident situations than at any other traffic conditions simulated. This conclusion is shown in summary The results from the ANOV A analysis as seen in Tables 3-21  improve the overall network conditions. CORSIM is not perfect, but researchers must continue to utilize the software in experiments so that feedback can be provided to programmers. As the software improves, it will become more advantageous and efficient to apply it to the evolving complex challenges that face transportation officials.

Recommendations
The recommendations that the research yielded are as follows: (1) Recommended Diversion Practice -It is recommended that RIDOT's traffic division and TMC review the diversion protocol, presented in Chapter 5 in Table 5-1, before deploying it to help manage the traffic problems related to un-planned minor traffic incidents.
(2) CORSIM as a Tool for Diversion Strategy Analysis -A well-developed and validated CORSIM simulation model can be used to aid in the development of a freeway-to-highway route diversion strategy. It is recommended that CORSIM be utilized in traffic diversion simulation applications.
(3) Application of Research Methodology -It is recommended that the summary research methodology developed be applied for similar simulation projects that utilize CORSIM or similar traffic simulation models.  In th e PI TI Car-Fol lowing Model . the basic assumption 1s that the foll ower vet11cle will try to mai ntain a space headway equal to For the current calculation for the follower vehicle , we are give n x. u, y, v and we must calculate a. Tl1e desired position at time t +Tis given by equati on (1 J as x, and v , v ·· aT and thus equation (2) becomes Note: Since th e term (u -v)2 is small . the approximation of v, = v 1 s used. Any difference 1s accounted for by th e calibration of b.
Solving for the acceleration of the followe r vehicle using equation (3) results in Equation (4) represents the basic car-following relationship . The term involving the constant b was introdu ced to allow fo r high re lative closing speed behavior obse rved empirically . The val ue of b has been calibrated to An emergency constraint overrides the car-following rules established above to prevent collisions . The basic concept provides that the follower vehicle can stop safely behind the leader vehi cle under the following conditions : Th e leader vehicle decelerates to a stop at the maximum emergency deceleration .
The follower vehicle . starting at the lag time c, later decelerates to a stop behind the leader vehicle at a deceleration rate within the maximum emergency deceleration limit.
If the leader stops at the maximum deceleration then u, = 0 and u2 The follower vehicle stopping at the maximum deceleration will also give Since the headway between the vehicles must exceed the length of the leader vehicle , equations (6) and (7) y'eld x, -Y, (u2 -v2) xy - The baste headway constrai nt th en becomes If x ,. u,. y, v, and Tare given . then the acceleration a of the follower vehicle for the time period (t, t + T) must be determined such that the headway constraint is not viola ted .
Two possible cases can arise : 1. The follower vehicle has a speed v , > 0 at tim e t+ T 2. The follower vehicle comes to a stop during the interval (t, t+ T). assuming th is occurs at time t (1 +p). where Substituting for v, and y , into equation (9) yields ec From equation s (10), (11) and (12) ( and v2 L x , The condition v, , / (u , 2 -e 2 c 2 ) -ec > o reduces to Equation (15) can be simplified to Equations (16) , (17), (18), and (19) are the constraints which determine the follower veh icle 's acceleration which must be maintained in order to satisfy the emergency noncoll ision conditions . Provided the vehicles are in a safe position at time t, then the above constant set will be sufficient for the vehicles at time t+ T. In particular B 2 +4C is always positive and thus the accele ration given by equation (16) has a real value .
The emergency constraint, however, is also used in the lane changing mechanism where the vehicles (i n adjacent lanes) may not be in a safe position relative to each other in a longitudinal sense . In this case the following can occur: 1) The above constraint set provides real acceleration but it is greater than e and thus the lane change is not initiated .
2) The discriminant (B 2 +4C ) is negative . In th is case the lane change is automatically not initia ted , sine the two veh icles must be in an unsafe relat ive position for occupying the same lane.

Summary of the Research Methodology
This appendix presents a concise summary of the research methodology in accordance with one of the four primary objectives of the research. Although the summarized methodology presented is minimal in procedural explanation, its substance should not be underestimated. This summary methodology is a framework outline that is intended to ensure that others perfom1ing similar modeling studies do not neglect any phases of model development or analysis.
The summary methodology was produced directly from the complete research methodology described in Chapter 3. It is presented as a set of six stages each containing a series of open-ended tasks. The tasks are presented as general actions, as opposed to set procedures, because each could be completed utilizing vanous modeling and analysis techniques. For more infomrntion on any stage or task presented, Chapter 3 of this document could be used as a reference.
Task I-1: Task I-2: Task I-3: The recommended deployment plan focuses on the ITS components required to deploy a route diversion strategy. By focusing on one diversion route strategy an ATMC could establish its role as an accurate information provider to the travelers using I-95 through the Providence Metropolitan Area (PMA). Once the people of RI and frequent travelers through the PMA realize the benefits of the information supplied through the visual ITS components, they will apply the actions recommended by the ATMC with greater confidence. Also, by concentrating on one area at the onset of the program, the incident management and response team can fine tune procedures before approaching other areas on the state's freeways and arterials. It is most important that the ITS infrastructure put in place for incident management be compatible with all regional equipment and any future ITS equipment RI utilizes.

STAGE-I DEVELOPING MODEL FOUNDATION
The National ITS Logical Architecture (NITSA) was developed using structured analysis techniques and consists of data flow diagrams, process specifications, and data dictionary entries. NITSA is designed to ensure that ITS projects are planned as flexible systems that will be compatable with regional and future ITS developments. In order to determine exactly which components and functions of ITS, the project should incorporate; the user benefits, or user services, must be defined onset of the system design. The user services represent the functions that the deployment is envisioned to provide to the travelers.

User Services
The following user services were identified using the NITSA: •!• Pre-Trip Travel Information

Physical Architecture
The Physical Architecture provides agencies with a physical representation (though not a detailed design) of the important ITS interfaces and major system components. It provides a high-level structure around the processes and data flows defined in the Logical Architecture. The principal elements in the Physical Architecture are the subsystems and architecture flows that connect these subsystems and tem1inators into an overall structure. A physical architecture takes the processes identified in the logical architecture and assigns them to subsystems. In addition, the data flows (also from the logical architecture) are grouped together into architecture flows. These architecture flows and their communication requirements define the interfaces required between subsystems, which form the basis for much of the ongoing standards work in the ITS program. A representative diagram of the physical architecture can be found in Figure  shows the physical components of the system that must be connected in some way by either wire line or wireless communications.

System Inventory
With the logical and physical architectures m place the actual system components can be specified. The system inventory will be presented in five subsections that represent the most important systems of the project deployment. The sub-sections have been divided as follows: (1) analysis system, (2) communication system, (3) detection, verification, and monitoring system, ( 4) advanced transportation information system (ATIS), and (5) advanced transportation management system.
Following the sub-sections is (1) Analysis System: Essentially: CORSIM, ITRAF, TSIS, and a Windows based computer.
However, as the detection, verification, and monitoring systems begin compiling data, the analysis system will expand to include these tools as model validation and calibration tools. The combination of the new data and the existing analysis system will make for more accurate traffic analysis in the future.
(2) Communication System: 262 The communication system is the backbone of any ITS deployment. Special consideration must be given to exactly what is installed and where because future ITS deployments will realize much greater benefits if they can utilize an adequate and established communication system. With future ITS projects in mind, it was decided that Tl fiber optic cable will be installed along approximately five miles of Rt. 10 and along six miles of I-95. This Tl line will be used to transmit large amounts of information to and from the A TMC in Providence. The CCTV cameras and loop detectors will be connected by coaxial cable and copper wire, respectively, to 6 multiplexers located roadside. The multiplexers will digitize and structure the information according to existing data-communication standards before transmitting it to the A TMC. Once at the ATMC the information will be decoded using another multiplexer and monitored. All cameras will constantly relay video images to the ATMC, however only images that the operator requests will be viewed on the display monitors. Four multiplexers will be located along I-95 and two will be located along Rt. 10. The multiplexers along I-95 will be stored at the base of the permanent DMS.
The Tl line will also transmit the data from the ATMC to be displayed on the DMS signs through the multiplexer at their respective bases and a short line of co-axial cable. Video switching equipment bas not been specified at this point because all but two cameras will constantly be displayed at the ATMC on one of the 15 small screen displays. Also connected to the multiplexers will be three HAR AM radio emitting beacons. These will also be connected using copper wire lines.
It will be very important to establish excellent emergency communications links as well. The emergency communications will begin with the eight roadside call 263 boxes. The roadside call boxes will be connected by copper wire line to multiplexers then onto the incident management center at the A TMC. In addition to the roadside call boxes an emergency cell phone number will be established and posted on signage as well as on the traffic management web site. Finally, three HAR broadcast beacons will be placed in a configuration so that their combined broadcast range will cover the entire PMA and approximately five miles to the south and for two miles north of the metropolitan line.
(3) Detection, Verification, and Monitoring System: The detection and monitoring systems allow the operators to constantly assess exactly how the system is functioning. The monitoring system consists of 17 surveillance CCTV video cameras, with pan, tilt and zoom capabilities, located along the mainline and alternate route. The detection and verification system will consist of 10 sets of loop detectors. CCTV and loop detection was chosen for this system because it is a proven technique that lends itself to the level of funding that was allotted for this project.
(4) Advanced Transportation Information System {ATIS): From a roadway system user standpoint, the advanced transportation information systems (A TIS) will present the most identifiable aspects of this project.
The A TIS must be strategically located and rigorously maintained. The appearance and locale of the equipment is directly related to the user's perspective of the value of the information provided. Any aspect of the ATIS that detracts from the information will encourage a learned disregard for the system and information. For these reasons 264 location, number, and connection to the ATMC have been given thorough consideration.
The ATIS consists of four permanent DMS, 8 flashing HAR signs; an HAR broadcast frequency, automated HAR message updates, 3 alternate route trailblazer signs, and a continuously updated traffic management web site.
(5) Advanced Transportation Central System : The central system is the ATMC. The A TMC is located in the heart of the PMA and takes in and disseminates all the information from and to the previously described systems.
Besides computing and display equipment, the ATMC is the place where human operators constantly monitor the network. The equipment that must be incorporated in the ATMC for this project includes: fax machines, Windows NT based computers, large screen and a matrix of 15 smaller screen video displays, a multiplexer, CODECS, telephones, internet, HAR message center and agency hot lines.
The most important feature of the central system is the responsive database software. This software will be developed using the results of the CORSIM analysis and the real-time information that is attained from the ITS. The database will consist of the range of messages to be played over HAR, the messages to be displayed on the DMS, and the links to posted on the web site. All this information will be determined based on manual input from the operator, based on the conditions they are witnessing, and the downstream and upstream occupancy determined by the inductive loop detectors. 265