Forecasting travel times based on actuated
`
`and historic data
`Danko A. Roozemond
`Delft University of Technology, Department of Civil Engineering,
`P.O. box 5048, Delft, The Nethelands
`E-mail: d.roozemond@ct.tudelft.nl
`
`Abstract
`
`In this research a model of forecasting the travel times of links will be addressed. Forecasting is
`one of the main topics of an integrated traffic management system and a necessity of dynamic
`route planning systems. To be able to forecast properly we use both historic data and current
`data from monitoring devices as input for our dynamic model. Thus, combining the best of
`both worlds, we are able to forecast travel times in the near future based largely on current data
`as well as travel times for some time ahead based on current and historic data. Accuracy and
`variability of data are important as they are the key element if these models will be
`incorporated in route guidance and traffic management schemes.
`
`1 Introduction
`
`Current estimates are that 65 percent of peak-hour travel on highways and urban
`roads and some ten percent of all daily urban travel is conducted under
`congested conditions [1]. In recent decades traffic problems have become both a
`social and economical embarrassment: congestion, deteriorating road safety,
`regression of mobility and environmental effects of traffic are widely considered
`important issues. The inability of the existing road network to cope with
`increased demand has been identified as one of the pressing infrastructure issues
`of this decade. Past custom to counter increased congestion with more, safer and
`wider roads, is currently giving way to more complex management and control
`systems and road pricing policies. Traffic management professionals and policy
`makers now turn to traffic management systems. The goals of traffic
`management systems are, among others: maximise safety and transport
`productivity; minimise congestion and damage through incidents; distribute
`information on traffic-conditions, road-conditions, weather, etc. A traffic
`management system should invoke appropriate intervening action when
`undesirable situations arise.
`
`As transportation is widely accepted as a crucial economic factor, a system of
`improvements to traffic management arises as the acceptable form to resolve the
`
`Transactions on the Built Environment vol 30, © 1997 WIT Press, www.witpress.com, ISSN 1743-3509
`
`Google Ex. 1019
`
`

`

`216 Urban Transport and the Environment for the 21st Century
`
`social and economic dependencies on transportation in the world. A more
`efficient use of existing infrastructure may gain time in which to develop
`alternate modes of transport and/or infrastructure. As with all demand-supply
`problems, solutions made through traffic management can be viewed as either
`increasing capacity to meet demand or modifying demand to levels deliverable
`under certain conditions. Both views rely as much on the actual transport
`conditions as on the perception of those conditions by the road user. A definite
`advantage can be gained by making the right kind of information available to the
`right kind of users. This requires the exchange of information between road
`system managers and the public.
`
`To handle the information requirements for route-guidance, traffic
`management and intelligent traffic control, one needs systems that accurately
`handle the behaviour of large complex road systems, while maintaining high
`levels of usability. To handle these information requirements one needs systems
`that accurately predict the, in essence unpredictable, behaviour of traffic
`participants. To be able to forecast we use both historic data and current data
`from monitoring devices as input for our dynamic model. Accuracy and
`variability of data are important as they are the key element if these models will
`be incorporated in traffic management schemes.
`
`2 Integrated Dynamic Traffic Management System
`
`To put IDTMS in perspective, integration of dynamic traffic management
`systems is no cure for all traffic problems, although it may streamline and reduce
`the traffic load and give better information and advise. Whatever the
`improvements, more direct routes available or more up-to-the-minute
`information, there would be a traffic problem, or new ones would arise.
`
`Benefits are not always easy to quantify as they are related to aspects like
`environment, information and finance and are difficult to quantify for groups or
`the whole system. The IDTMS is more than a mere management system,
`although operational benefits rely heavily on the forecasting properties of the
`system and the ability to act and adapt dynamically.
`
`2.1 Decision making in an IDTMS
`
`Traffic management can be divided into three different levels of decision making
`each with its own data demands [2]:
`
`• short term traffic control decisions; traffic management with a time-frame of a
`few minutes, implemented in traffic control systems; in most cases no human
`action is required.
`• medium term traffic management; based on a few hours time-frame,
`implemented in traffic management systems.
`• long term planning; based on a large (days to years) time scale, also called
`traffic planning systems used in a highly interactive environment as a tool to
`human design and planning efforts.
`
`The different levels of decision making have consequences for the required
`data. Long term planning can be done with aggregated, semi-dynamic data and
`is not used by travellers, but rather by policy makers and planners. Much more
`detailed data, more accuracy, reliability and actuality, is required for medium
`
`Transactions on the Built Environment vol 30, © 1997 WIT Press, www.witpress.com, ISSN 1743-3509
`
`Google Ex. 1019
`
`

`

`Urban Transport and the Environment for the 21st Century 217
`
`term planning. Highly accurate, up-to-date data is necessary for short term
`decisions in traffic control. To get the required data, on-line monitoring of traffic
`conditions is crucial for traffic control, but also very valuable for traffic
`management applications and profitable for planning purposes.
`
`2.2 Architecture of an IDTMS
`
`One of the main overall objectives of this project is to develop a framework
`integrating traffic and infrastructure planning, traffic management, traffic
`control, information and simulation systems into one multi-user, multi-discipline
`Integrated Dynamic Traffic Management System (IDTMS). To satisfy that
`objective, a dynamic traffic management system should be based on modular
`and distributed components that can operate within an open, evolutionary,
`distributed and scalable architectural framework serving the needs of several
`management layers [3, 4]. Standardisation is a requirement for components
`within the system to facilitate access to and manipulation of the information and
`models present in the system. Sub-systems are designed to perform as
`autonomously as possible, co-ordinating their own actions, interacting with
`other sub-systems trough standardised interfaces when necessary. This
`increases overall robustness of the system and creates an appropriate
`environment for information and traffic management systems that can address
`several goals and even can handle different and opposite goals. Forecasting is
`one of the main topics of an IDTMS and a necessity of dynamic route planning
`systems.
`
`When focusing on integrated network control, forecasting, incident detection
`and re-routing systems, there is a need for systems that accurately handle the
`large quantities of data and dependable data collection. In order to be effective,
`such systems need to collect and maintain data locally for immediate use, and on
`request aggregate data before it is presented to other components via a data
`handling system. For data handling, databases based on a GIS are proposed as
`one of the possible solutions. The GIS will perform the function of underlying
`database (storage, handling, etc.) and query language. Uniform standards
`should apply to any particular service in the overall system. What we need is a
`scalable system for shared use of distributed data that will search for the data,
`after merging available data from different sources, generate (un)available values
`or transpose the data and present it. The R&D efforts on distributed computing
`can be exemplified by a number of projects. The Andrew File System (AFS) is
`an ongoing project on distributed file systems, creating a virtual single file
`system with potential to provide terra-bytes of data on-line [5]. Project Athena
`recently ended in a campus wide distributed computing environment supplying
`services from and across networked resources by virtualising the classical link
`between node and user [6]. The wide spread of data across different
`organisations and installations within organisations creates a very practical
`problem for users and developers of modelling software for data intensive
`engineering applications. The obvious advantages of shared use of data are
`dwarfed by the problems generated when restricted use, data integrity, access
`security and cross accounting has to be considered. Not only do these pose a
`problem in general, but especially when large volumes of data have to be
`gathered in order to satisfy a single request, the system overhead to handle all
`details can easily explode.
`
`Transactions on the Built Environment vol 30, © 1997 WIT Press, www.witpress.com, ISSN 1743-3509
`
`Google Ex. 1019
`
`

`

`218 Urban Transport and the Environment for the 21st Century
`
`The data handling system should be able to deal with the large volumes of
`distributed data generated and consumed by modelling, simulation and DSS
`applications common to integrated traffic management systems. These problems
`will have to be addressed in a single comprehensive system, making use of
`existing distributed sources of scientific data sets [7]. Specific problems dealt
`with are the following:
`• access to and storage of time series data and spatial grid data independent of
`storage format and storage location;
`• storage and context mapping of data generated, making it available for later
`retrieval using the same system;
`• extending and transforming data to specified interval or grid, delivery of data
`in a format suitable for the requesting application.
`
`3 Dynamics in traffic systems
`
`As we all know: traffic is not fully predictable. Given all the initial facts, it is not
`always certain that accurate predictions can be made about the future; not
`everything is fully deterministic and non linearity is intrinsic to the system. This
`raises the question of how to avoid getting into an uncontrollable chaotic
`situation, which is not that functional for traffic control and traffic management
`systems; or when found in the middle of it, how get out of this situation with as
`many positive side-effects possible. Concerning traffic management and control
`we are in need of a theory that can cope with this uncertainty and complexity.
`From a mathematical point of view the stability and behaviour of participants in
`transportation systems are essential parts of the system. Traffic management and
`information provision for minimising congestion is made difficult not only by
`the magnitude of the problem but by the diversity of the interacting "intelligent
`agents" whether persons, (automated) control objects, etc. An intelligent
`transportation system must not only handle real-time needs; it must also be a
`system that adapt to changing system parameters and structures, continually
`improving its ability to act as well as react.
`
`Traffic management and traffic control in cities is the most problematic and
`complicated the whole transportation system as well as the most challenging and
`least researched. In urban areas the complexity exceeds that of the highway
`conditions due to the variety in means of transport, the difference in the speed of
`the participant and the greater probability of unanticipated events. Influences on
`the instability in highway conditions are for example geometry of the road, types
`of vehicles, motive for the trip, weather conditions and light conditions.
`Furthermore the driver itself is a great source of instabilities.
`
`As traffic management systems must react to the different states of traffic flow in
`the network the management system is the actor controlling and reacting to
`dynamic actions in the system. To be able to do so, a fully pro-active dynamic
`control mechanism should be an integral part of the IDTMS. For a more
`elaborate overview of adaptive traffic management we recommend Cuena et al
`[8]. To satisfy the information requirements for route guidance, traffic
`management and intelligent traffic control, one needs systems that cope with and
`predict the dynamic behaviour of large complex road systems. And, as the aim is
`to predict and prevent unwanted situations in the transport system there is an
`obvious need for traffic forecasting models on all time scales and the most
`wanted are the short- and medium term forecasting models.
`
`Transactions on the Built Environment vol 30, © 1997 WIT Press, www.witpress.com, ISSN 1743-3509
`
`Google Ex. 1019
`
`

`

`Urban Transport and the Environment for the 21st Century 219
`
`4 Forecasting dynamic transport systems
`
`As the reasons for forecasting may be obvious, the real advantages are viable
`when there are applications become available that do something with these
`models. We can see this happen in the next decade as more and more in-car
`equipment becomes available and humans are getting used to. As congestion
`levels grow and opposition to new roads grow also, the temporarily solution is
`to get more people over the same road. When people (traffic managers and the
`road users) get the forecasting information they are, given the information
`content is useful, using that information. Dynamic route guidance systems, with
`integrated forecasting properties, could provide significant benefits to both
`individual drivers as well as the overall transportation system. The effectiveness
`of systems that provide traffic information and their potential for reducing
`congestion, depends heavily on drivers' reaction to additional information.
`Drivers would avoid traffic congestion and incidents, and (theoretically, at least)
`arrive at their destination as fast as possible. Utilisation of the roadway system
`would improve, because many drivers could (theoretically, again) avoid already
`congested thoroughfares and lessen the severity of existing congestion. Usually
`traffic participants can choose between the different modes of transport. Given
`enough information the preferred mode of transport may change.
`
`4.1 Travel time forecasting
`
`Given a road map, the possible roads can be represented by several sections,
`each with their own characteristics. Every section has an entrance and an exit
`gate; the traversing time can be calculated with historic data. Every road sub-
`system has the underlying smaller-road-set that will only be used for transport
`inside the section and adjoining sections. Several specified road segments can be
`combined to a traversing link, thus creating a higher level for non local traffic.
`Example: Travelling from origin to a destination in an other city; the system start
`using the most detailed network with all routes bottom-up wards, until you are
`on a tertiary road-network; from there you are directed to travel further upwards
`until you use the main road-net (highways). Coming close to the destination the
`system uses the road-networks from the high level down.
`
`The basis for travel time forecasting is not that difficult: if a person can address
`the travel times on all possible routes and links, one can calculate the overall
`travel time of a proposed journey. As mentioned earlier this is only true for
`predictable situations and in most cases this isn't the case in traffic forecasting
`nowadays. The kind of calculus used is effectively a state prediction model:
`predicting the states on time tj. Every road segment has its travel time prediction,
`based on historic data, at a certain time or time span.
`
`For the several possible routes from origin to destination the chosen objective
`(time) should be calculated by formula's like [3]:
`
`total travel time = ]T A^.% • £.^ , where
`All links
`j — Start Node of link 7, k
`k = End Node of linkj.k
`Afy* = Traveltime on link;,/:
`£\ ^ = Correctionfactor for link y, k
`
`(1)
`
`Transactions on the Built Environment vol 30, © 1997 WIT Press, www.witpress.com, ISSN 1743-3509
`
`Google Ex. 1019
`
`

`

`220 Urban Transport and the Environment for the 21st Century
`
`Travel times consisting of several adjoining links can be calculated easily by
`calculating the solely links and adding them to one total trip-time. It may be clear
`that this kind of model is not that accurate as none of the disturbances that make
`traffic prediction so awkward are included. So there is a need for a more
`elaborate model end preferably a dynamic one.
`
`4.2 Dynamic forecasting based on actuated data
`
`While static systems are entirely self-contained, dynamic systems require
`accurate and up-to-date traffic data - data that must be provided by agencies
`through some type of infrastructure based monitoring device to user
`communication scheme.
`
`To be able to forecast properly we use both historic data and current data
`from monitoring devices as input for our dynamic model. Thus, combining the
`best of both worlds, we are able to forecast travel times in the near future based
`largely on actuated data as well as travel times for some time ahead (>15
`minutes) based on actuated and historic data. In general we can see that the need
`for actuated data becomes smaller as the needed forecast lies further away.
`Accuracy and variability of data are important as they are the key element when
`these models will be incorporated in traffic management schema's.
`
`The needed data itself needs to be extracted from the system; only specified data
`is given to the forecasting system:
`•
`the relevant road network given the links to be visited (origin-destination
`or preferred route);
`additional data on important aspects of traffic conditions: kind of day,
`part of the day, weather conditions, road conditions, etc.;
`additional data on important aspects of driver/car conditions: kind of
`driver: behaviour, age, etc., kind of car: velocity, acceleration and
`deceleration limits, etc.
`
`•
`
`•
`
`The forecasting system gets its data specified as road segments; as an object
`that 'knows' the forecasting equations, actual traffic conditions and constraints
`and, in combination with the above given data, future situations can be
`calculated. If the desired information is not in the near future, this can be
`simulated or calculated as accurate as possible with historic data.
`
`The kind of calculus used is effectively a state prediction model: predicting
`the states on time tj, given the state on time tj-dt-(where dt is not fixed). Every
`consecutive road segment gives its calculated travel-prediction, based on
`actuated and historic data, at a certain time given all the different inputs and the
`whole trip can be calculated. The travel time of a certain link is calculated on the
`moment that, real or virtual, car comes from the adjacent link, node or gate and
`enters the new link. So the correct entering time-state is known or, in case of a
`virtual car, calculated from the exit gate of the previous link.
`
`The whole optimisation process for the objective travel time can basically be
`written in a few formula's (2):
`
`total travel time =
`
`Ar.^^ • £.^^ , where
`
`(2)
`
`Transactions on the Built Environment vol 30, © 1997 WIT Press, www.witpress.com, ISSN 1743-3509
`
`Google Ex. 1019
`
`

`

`Urban Transport and the Environment for the 21st Century 221
`
`j = Start Node of linkj.k
`k = End Node of Ymkj,k
`
`T = Time
`Atj^j = Traveltime on link;,/: at time = T
`
`£j,k,T ~ Correctionfactor for \inkj,k at time = T
`
`(2a)
`
`Ar^j = (/^,r/^,r )^«driver/car , where
`Ij^j = Length of link j,k at time = T
`Vj,k,T - Mean velocity on link j,k at time = T
`^driver/car = non linear term on car and driver affecting traveltime on link j,k
`W,...) , where
`
`^ '
`
`%j^(t) = factor for congestion level at time t
`
`Note : The actual implementation could be far more complex then these formula's indicate
`as the non-linear terms should be specified, several inputs will be calculated via fuzzy set
`theory and the specific factors are further specified.
`
`The predictability of the mean velocity on link j, k at time T is the key
`element. Aggregated historic data can be used in cases when no direct time T is
`given or when T minus current time is larger then a couple of hours. In that case
`no specific extra information is in the actuated data. There is however extra
`information in weather forecasts, etc. So these should be used if available to
`make a better prediction. For time-spans less the an hour, especial when smaller
`then 15 minutes, a forecast based on actuated data is more accurate. In
`intermediate cases a fuzzy decision making process could determine what kind
`of mix of actuated and historic data should be used and what variables should be
`used. Fuzzy logic and fuzzy set theory provide a rich and meaningful addition to
`standard logic. Fuzzy quantifiers don't give the count exactly, but fuzzily in that
`way it can deal with fuzzy probability like 'not very likely', 'rarely, or 'fairly
`possible' [9].
`
`Even when predicted accurate the forecast is only valid for a limited time. In a
`couple of minutes the whole situation can be different. Uncertainty is present in
`most tasks that involve humans. In this case the prediction times will be
`uncertain, especially for trips longer then approximately one hour. In the
`meantime travel situations may improve or get worse.
`
`The main problem is that an a priori solution for the entire planning horizon
`ignores the unpredicted and unpredictable real-time variations of non recurrent
`variations in network characteristics. A solution to this problem, also
`implemented in the above given model is the use of a rolling horizon approach
`[10]. It addresses the scenario in which demands are not known a priori for the
`entire forecasting horizon. Long-term forecasts of future traffic conditions may
`not be as reliable as short-term forecasts. In rolling horizon approach the
`forecast horizon is subdivided into several stages or projection horizons with a
`limited time span; in each stage, reliable short-term forecasts and not-as-reliable
`medium- and long-term forecasts of traffic conditions are available when each
`user enters that stage. When solutions like these are used a more optimal route
`can be calculated during trips. An in-car device then becomes a necessity.
`
`Transactions on the Built Environment vol 30, © 1997 WIT Press, www.witpress.com, ISSN 1743-3509
`
`Google Ex. 1019
`
`

`

`222 Urban Transport and the Environment for the 21st Century
`
`It is presumed that data on upstream traffic can be used to predict the
`downstream traffic in a specific time period. Research has shown that models
`using historic, up-stream and current link traffic give the best results [12]. These
`models only give forecasts for specific links and don't take other links, possible
`disturbing that link's traffic into account. Thus a more elaborate model could
`give better results and for route guidance such a model could be more practical.
`The model used in that study [12] was (3):
`
`Z/+i = forecasts at time t + 1 on link i
`t//** = observed traffic at time t + lon link i + k (origin link)
`'+*-i) _ observed traffic at time t + lon link i + (k - 1) (adjecent link to origin link)
`+(k+i) _ observed traffic at time t + lon link i + (k + 1) (adjecent link to origin link)
`C\ = current traffic at time t on link i
`C/_j = current traffic at time t - 1 on link i
`C/_2 = current traffic at time t - 2 on link i
`C/_3 = current traffic at time t - 3 on link i
`yH}+i = historical average at time t on link i
`jp, /are weighting parameters
`
`Several other traffic forecasting methods can be used or even incorporated in
`the system. One of the most promising is the use of Artificial Neural Networks
`(ANN) for traffic forecasting [11]. The ANN techniques are notable for their use
`in addressing non-linear problem in data rich environments. The findings in
`most of the cases are that predictions, in the range till 15 minutes, are better then
`in ordinary cases. So the general idea is that predictions less then 15 minutes are
`viable by several means, ANN or general mathematics. The great virtue of the
`mathematics approach is that all inputs can be monitored and the influence on the
`outcome can de mathematically stated as in neural networks the total set of inputs
`gives an output, and you haven't got the slightest idea in what way the model is
`operating. So gaining inside in forecasting can best be done with (non-linear)
`mathematic models.
`
`We have not discussed the actual calculation part in depth, but in practise the
`calculation is the optimisation of a directed network. The optimisation techniques
`used are OR techniques. Faster algorithms, suitable for distributed computing
`can be useful as brute force is not that elegant and takes quite a long time.
`
`5 Dynamic route guidance
`
`Dynamic route guidance systems are the systems in traffic engineering relying
`heavily on forecasting models for accurate predictions. Fully dynamic route
`guidance systems offer additional benefits over semi-dynamic ones, in that they
`can be responsive to random fluctuations of actual traffic conditions. Those
`random fluctuations may be due to accidents or just some fluctuations in actual
`traffic conditions. The common feature of those random effects is that there is no
`way to predict them from historic data alone. In case of an accident, traffic can
`be redirected to avoid congestion through a new optimum route. One specific
`function of a dynamic route guidance system, of particular interest, concerns
`route choose calculations. The algorithm used to perform this task is the basis of
`the success of this part of the route guidance system; response time,
`
`Transactions on the Built Environment vol 30, © 1997 WIT Press, www.witpress.com, ISSN 1743-3509
`
`Google Ex. 1019
`
`

`

`Urban Transport and the Environment for the 21st Century 223
`
`performance, effectiveness and detailedness are all important. Criteria for the
`"best" route are: shortest, cheapest, fastest, prettiest, etc.
`
`For time dependent route planning algorithms dynamic and real time
`implementation is a necessity to reflect changes in traffic conditions. When a
`large number of cars have the same data and the same linear rules a situation can
`occur where an all or nothing assignment gets real. We are thus concerned with
`modelling dynamic networks when drivers simultaneously optimise their
`departure time and route choice. We state equilibrium conditions and propose a
`simulation-based model that can solve large networks accounting for all realities
`of actual networks. An approach based on user behaviour rules or statistics
`should be used to get a dispersion of the traffic flow and guide traffic to a less
`congested roadway. We need a time-dependent optimum path algorithm that
`computes paths on multi-modal networks, accounting for mode switching costs
`(time and money).
`
`The best place for route planning algorithms to operate is in each vehicle. The
`vehicle knows its progress and is able to fuse the actuated data with historic data
`and other sources and sending the vehicle data on travel time estimates. If there
`is two-way communication the vehicle may also be used as a probe for
`determining travel times on traversed links.
`
`6 Further research and conclusions
`
`An IDTMS is no cure for the traffic problems; it may streamline and reduce the
`traffic load, have more direct routes available, give good information and advise
`and still there would be a traffic problem. So research on changing attitudes,
`other ways of transport, etc. is equally important. An integrated design of an
`IDTMS that will easily include all mentioned options has yet to be developed.
`Seen in a broader perspective, allow for a sensible traffic management strategy
`to be executed by interconnected, modular intelligent traffic control systems.
`Switching to these new systems will, in the long run, enable cheap maintenance
`and easy transition to arising new facilities in traffic management based on the
`given infrastructure. To be able to economically develop IDTMS in the future,
`adopting rigorous interfacing and networking standards and implementing these
`in current investments is necessary now. To be able to even partly solve
`planning, management and control problems, there is a definite need for more
`suited dynamic models and control strategies.
`
`Traffic forecasting on historic data alone doesn't give a good enough basis for
`traffic forecasting in congested areas. In case of congestion a better forecasting
`model could be build using current data on that specific link, historic data (from
`similar time and road conditions on that specific link) and current data from
`adjoining links. At this stage we are not certain whether this forecasting principle
`is indeed a working solution, but the expected and very preliminary results seem
`to be very promising.
`
`The dynamic behaviour of traffic participants will change if information based
`on actuated data is presented. The specific changes and its influence on traffic is
`not yet clear and the influences of data availability is one of the remaining
`questions. For traffic control the research agenda comprises to adjusting the
`control schemes to appropriate ones that can deal with dynamic and actuated
`
`Transactions on the Built Environment vol 30, © 1997 WIT Press, www.witpress.com, ISSN 1743-3509
`
`Google Ex. 1019
`
`

`

`224 Urban Transport and the Environment for the 21st Century
`
`data. Some combination of historic data and actuated data will be needed to give
`predictions further ahead then the next 15 minutes.
`
`The subject of finding alternatives to the best path in a dynamic environment
`based on dynamic traffic forecasting is not fully functional yet. It remains a
`question to what extend this will be operational in the (near future) and effective
`implemented in route guidance systems. We should recognise that drivers may
`want to evaluate paths based on a combination of travel times and direct costs.
`While this is straight forward when these functions are linear, it is much more
`difficult when this is not the case. New algorithms, to efficiently solve these
`non-linear problems, are required.
`
`REFERENCES
`
`[1] Bernstein, D. (1993), Congestion pricing with tolls and subsidies, in:
`proc. Pacific Rim TransTech Conference, voi 11, ASCE, New York
`[2] Romph, E. de (1994), A dynamic traffic assignment model: theory and
`applications, DUT-press, Delft
`[3] Roozemond, D.A. (1995), An Integrated Multi-Discipline Dynamic Traffic
`Management System, Based on Information, Objects and Inter-Object
`Communication; Urban transport and the environment in the 21st century,
`Sucharov, L.J.(ed.), CMP, Southampton
`[4] Gollu, A., Deshpande, A, Hingorani, P., Varaiya, P. (1994), SmartDb:
`an object-oriented simulation framework for intelligent vehicles and
`highway systems, in proceedings of 5th conference on A/, simulation, and
`planning in high autonomy systems, IEEE, Los Alamos
`[6] Champine G.A., 1991, MIT Project Athena : a model for distributed
`campus computing, Digital Press, USA
`[6] Mauro, T., 1989, An overview of the AFS, Transarc Corporation,
`Pittsburg
`[7] Johanns, R.D and Roozemond, D.A.,1995, JGI coping with distributed
`incomplete data, in Critical review of the applications of advanced
`technologies (proceedings of the fifth EuroplA international conference),
`Miramond, M. & le Gauffre, P. & Beheshti, R.& Zreik, K. (eds),
`europia productions, Paris
`[8] Cuena, J., Hernandez, J., Molina, M. (1995), Knowledge-based models
`for adaptive traffic management systems i\\:transpn. res.-C, vol3, no 5,
`Elsevier
`[9] Zadeh, L.A. (1987), The role of fuzzy logic in the management of
`uncertainty in expert systems, in: Fuzzy sets and applications: Selected
`papers by L.A. Zadeh, Yager, R.R., Ovchinnikov, S., Nguyen,
`H.T.(eds), John Wiley, New York
`[10] Peeta, S.& Mahmassani, H.S. (1995), Multiple user classes real-time
`traffic assignment for online operations: a rolling horizon solution
`framework, in: Transp Res-C, vo!3, no 2, Elsevier
`[11] Lyons, G.D., McDonald, M., Hounsell, N.B., William, B., Cheese, J.,
`Radia, B. (1995), Urban traffic management: the viability of short term
`congestion forecasting using
`artificial neural networks;
`HTTP://www.smithsys.co.uk/ smith/public/tech/viabilit.html
`[12] Hobeika, A.G. & Kim, C.K. (1994), Traffic flow prediction systems
`based on upstream traffic; in: 7994 vehicle navigation & information
`systems conference proceedings ; IEEE
`
`Transactions on the Built Environme