Scholarly article on topic 'The need to combine different traffic modelling levels for effectively tackling large-scale projects adding a hybrid meso/micro approach'

The need to combine different traffic modelling levels for effectively tackling large-scale projects adding a hybrid meso/micro approach Academic research paper on "Civil engineering"

CC BY-NC-ND
0
0
Share paper
OECD Field of science
Keywords
{"Integrated framework" / "Traffic Simulation" / Aimsun / Hybrid}

Abstract of research paper on Civil engineering, author of scientific article — Jordi Casas, Josep Perarnau, Alex Torday

Abstract The evaluation of ATMS requires a network-wide assessment of their impact as opposed to an isolated analysis of key intersections. Microscopic traffic simulation models emulate the dynamics of individual vehicles in a detailed network representation, based on car-following, lane changing, etc; include route choice models to implement dynamic traffic assignment. Microscopic models are usually appropriate for operational analysis due the detail of information provided by the simulator. However, they are data intensive and have a significant computational cost. Mesoscopic models combine simplified flow dynamics with explicit treatment of interrupted flows at intersections and allow modeling of large networks with high computational efficiency. However, the loss of realism implied by a mesoscopic model makes it necessary to emulate detailed outputs (detector measurements, instantaneous emissions). The above give rise to the need to combine meso and micro approaches into new hybrid traffic simulators where very large-scale networks are modeled mesoscopically and areas of complex interactions benefit from the finer detail of microscopic simulation.This paper presents the new system architecture implemented in Aimsun of the hybrid model with special emphasis on the consistency of the transition between the mesoscopic and the microscopic models, and vice versa and aims to demonstrate how consistency is complemented using computational results in a large-size network.

Academic research paper on topic "The need to combine different traffic modelling levels for effectively tackling large-scale projects adding a hybrid meso/micro approach"

Available online at www.sciencedirect.com

ScienceDirect

Procedia Social and Behavioral Sciences 20 (2011) 251-262

14th EWGT & 26th MEC & 1st RH

The need to combine different traffic modelling levels for effectively tackling large-scale projects adding a hybrid meso/micro

approach

Jordi Casasa'b'*, Josep Perarnaua, Alex Tordaya

aTSS- Transport Simulation Systems, Passeig de Gracia 12, Barcelona 08007, Spain bUniversity of Vic, Sagrada Familia 7, Vic 08500, Spain

Abstract

The evaluation of ATMS requires a network-wide assessment of their impact as opposed to an isolated analysis of key intersections. Microscopic traffic simulation models emulate the dynamics of individual vehicles in a detailed network representation, based on car-following, lane changing, etc; include route choice models to implement dynamic traffic assignment. Microscopic models are usually appropriate for operational analysis due the detail of information provided by the simulator. However, they are data intensive and have a significant computational cost. Mesoscopic models combine simplified flow dynamics with explicit treatment of interrupted flows at intersections and allow modeling of large networks with high computational efficiency. However, the loss of realism implied by a mesoscopic model makes it necessary to emulate detailed outputs (detector measurements, instantaneous emissions). The above give rise to the need to combine meso and micro approaches into new hybrid traffic simulators where very large-scale networks are modeled mesoscopically and areas of complex interactions benefit from the finer detail of microscopic simulation.

This paper presents the new system architecture implemented in Aimsun of the hybrid model with special emphasis on the consistency of the transition between the mesoscopic and the microscopic models, and vice versa and aims to demonstrate how consistency is complemented using computational results in a large-size network.

© 2011 Published by Elsevier Ltd. Selection and/or peer-review under responsibility of the Organizing Committee. Keywords: Integrated framework, Traffic Simulation, Aimsun, Hybrid

1. Introduction

The advent of Intelligent Transport Systems (ITS) has raised new challenges in terms of traffic modeling which have drawn the attention on traffic simulation fostering the interest for the development of improved microscopic approaches which emulate the dynamics of individual vehicles in a detailed network representation, based on car-following, lane changing, gap acceptance models consistent with traffic flow theory, route choice models to implement a dynamic traffic assignment, account explicitly with traffic control, appropriate for operational analysis

* Corresponding author. Tel.: +34 650395787. E-mail address: casas@aimsun.com.

1877-0428 © 2011 Published by Elsevier Ltd. Selection and/or peer-review under responsibility of the Organizing Committee doi:10.1016/j.sbspro.2011.08.031

and mesoscopic models which combine a simplified flow dynamics with explicit treatment of interrupted flows at intersections and allow modeling large networks with high computational efficiency dealing with dynamics aspects in strategic planning (such as DYNASMART, Jayakrisham et al. 1994, DYNAMIT, Ben-Akiva et al 2002, etc). The application to large complex systems has led to the necessity of using in an integrated way various simulation approaches. Barceló et al (2005) address this necessity from a methodological point of view, ratifie d from a practical point of view by Alexiadis (2007). A specially interesting case is when the analysis is done in the framework of traffic management focused on the study of the subarea related to the rest of the network, for instance to account for diversions of traffic flows around the subarea consequence of congestion or incidents in the subarea. In this case at a given time during the simulation it may happen that a fraction of the trips with origins and destinations outside the subarea that initially were traveling in paths across the subarea will be diverted to paths bypassing the subarea as illustrated in Figure 1. A way of dealing with this situation is simulating mesoscopically the whole network and microscopically the subarea network.

Figure 1 Combining Meso and Micro approaches

A vehicle traveling from origin i to destination j is continuously being tracked by the simulation engine, it follows a route from the origin until the border of the subarea according to the proposed mesoscopic simulation of traffic flows, after crossing the border the simulation engine changes the simulation mode to the microscopic approach along the part of the path in the subarea and when the vehicle completes the trip across the subarea arriving to the opposite border the simulation mode changes back to mesoscopic until it reaches its destination. In case of congestion arising in the area alternative paths around the subarea can become more attractive as shown in Figure 1. This combination of simulation approaches raises strong consistency problems when the Mesoscopic simulator and the microscopic simulator are two independent pieces of software. The integration of Meso and Micro simulators has been analyzed in detail by Burghout (2004), and Burghout et al. (2005), in the case of MIME and MITSIM. In particular the integration has to solve the consistency problems in route choice and network representation, traffic dynamics at meso-micro boundaries, traffic performance for micro and meso models and communication and data exchanges. Further than the network representation consistency problems other problems arise from the fact that vehicles in the meso model and vehicles in the micro model are not the same, and the path calculation is different. Barcelo et al. (2006b) and Casas et al (2010) propose a conceptual approach to overcome these drawbacks. This paper extends that conceptual approach, discusses its practical implementation and provides preliminary results.

The remainder of the paper is organized as follows. Section 2 discusses important requirements for hybrid models of high fidelity and presents a general framework for their implementation. Section 3 looks into the relevance of having an integrated framework and the details of the proposed hybrid framework. Section 4 discusses issues related to consistency of traffic dynamics meso-micro adding details how to deal with the boundary conditions and the global look ahead element as a key point detected for an hybrid model. Finally section 5 shows some experiment results, considering the global look ahead issue as a key point and section 6 concludes the paper.

2. Requirements for developing reliable hybrid models

The main requirements that the integration of micro/meso models needs to satisfy in order to develop reliable hybrid models include (Burghout and Koutsopoulos, 2006):

1. Consistency in network representation.

2. Consistency in route choice representation.

3. Consistency of traffic dynamics at meso-micro boundaries.

4. Consistency in traffic performance for meso and micro submodels.

5. Transparent communication and data exchanges

The requirements 1, 2 and 5 are guaranteed considering the integrated architecture implemented in Aimsun, explained in more details in section 3 and the requirements 3 and 4 are object of detailed study in this paper in section 4.

3. Integrated architecture and Hybrid framework

The way the integration has been implemented in Aimsun rests on three main pillars: (1) an object oriented data base that contains all the information forms both the demand (mainly OD matrices per vehicle type and time periods and Public Transport(PT) schedules) and the supply (Road infrastructure, traffic management actions, incidents, traffic signals, PT lines, etc.), (2) three different network loading models (macroscopic, mesoscopic and microscopic) and (3) three traffic assignment techniques, one static and two dynamic. Such structure is summarized in Figure 2.

Figure 2 Aimsun internal architecture

The object oriented data base will therefore contain all the information needed to feed the network loading and assignment processes. Each entity (road section, node, turn, Variable Message Signs, controller, etc) is described through different attributes (in the same way as a Geographic Information System). Some of the attributes will be used only by a specific model while others will be shared by all of the models. In this last case, speed limit is a perfect example: such road section attribute is used by the three network loading models. Figure 3 details this concept of the shared data base.

Figure 3 Object oriented data base

Concerning the two other elements which are the network loading and traffic assignment, the authors want to emphasize here that, given the adopted data architecture (e.g. sharing the same network representation), there is no need to tie any dynamic traffic assignment (DTA) to a network loading process. The intention of this comment is to clarify a common misunderstanding in the modelling community which is the habit of linking DTA with a meso network loading which is an overly restrictive vision of what an integrated framework can offer.

Indeed, DTA and even static traffic assignment are fed by travel times calculated by a model, whatever this model might be. The only difference, from a technical point of view, between the static assignment and the dynamic one is that the latter is time-dependent and produces various sets of paths and path flows, one per time period. One of the clear advantages of using a common road network representation is that traffic assignment results produced by any type of network loading modelling can be stored and reused for another simulation run , without having to apply the same model that was used to produce these paths.

From a practical point of view (the assignment results data flow capability is summarized in Figure 5), this architecture allows running scenarios such as:

• Running a macro static assignment

• Using these results to start a Dynamic User Equilibrium (DUE) assignment process with meso

• Using the results from the meso DUE for a microsimulation in which the Stochastic Route Choice only applies to informed (VMS, radio, navigation system) vehicles (Barcelo and Casas, 2006)

This assignment results data flow capability is summarized in FIGURE 5. The separation between Dynamic Traffic.

Assignment module and network loading modules is, in fact, a fundamental criterion for hybrid simulation consisting of running simultaneously microscopic and mesoscopic network loading, each technique being applied in a different part of the network and each one implementing their specific behavioural models (car-following, lane changing and gap acceptance models). The Computational Framework for implementing the hybrid meso-microsimulation framework was presented in (Casas et al, 2010), where the additional modules included in this framework was the META Event Oriented Simulator and the Vehicle Manager module.

The requirements for developing reliable hybrid models in the presented in section 2 satisfied taking into account the unique network representation and the integrated framework architecture are:

• Consistency in network representation: In the integrated platform, explained above, both models share a unique network representation, which means each model has its specific view of the same object in the network. As a consequence of this common representation, this consistency requirement is always satisfied because there is a unique network representation. However, each submodel (meso and micro) has an internal representation in order to model the traffic dynamic in the links.

Consistency in route choice representation: The integrated model architecture (see Figure 5) guarantees consistency in route choice representation because the dynamic traffic assignment server module has a unique route choice representation, independent of whether the links are defined as mesoscopic or microscopic. Transparent communication and data exchanges: The exchange of information is carried out by the Vehicle Manager module, which has a unique representation of each vehicle and is shared by both submodels. Regarding communication, the synchronization of the two submodels is managed by the META Event Oriented Simulator.

Figure 4 Integration of the Dynamic Traffic Assignment Server and the Hybrid Network Loading

Figure 5 Paths assignment and OD matrices data flow chart

4. Consistency traffic dynamics meso-micro

The remaining requirements presented in section 2, and not covered by the integrated framework explained in the

previous section, are:

• Consistency in traffic performance for meso and micro submodels: The two submodels need to be consistent with each other with regard to the results they produce. Ideally, for those facilities that can be simulated sufficiently well by both models, the results, in terms of common outputs, such as travel times, flows, speeds, densities, etc, should be similar. This consistency is determined by the correct calibration of each submodel and establish the consistent simulation outputs of both models. In order to guarantee the consistency of both traffic models, a set of experiments were designed where the objective is the comparison of both simulation outputs. The different experiments contains different network topologies (urban, non-urban and mixed) and different traffic demand profiles. Figure 6 depicts the consistency in terms of speed-flow diagram of both models considering a highway.

• Consistency of traffic dynamics at meso-micro boundaries: This consistency has to be ensured in order to guarantee the transition between the meso and micro submodels has a continuity. For instance, when a queue is forming downstream the boundary point, and grows until this point, the queue should continue in the other submodel, upstream the boundary. by the vehicle manager module that transfers the boundary conditions between the mesoscopic and microscopic in all sections in the border.

Figure 6 Micro and meso Speed-Flow diagrams

4.1. Boundary Conditions

Consistency with respect to traffic dynamics at the boundaries of the two models is critical. Taking into account that both meso and micro models are vehicle-based and lane-based traffic flows one aspect that could be avoided is the aggregation/disaggregation issues between a non-vehicle-based to/from a vehicle-based and non-lane-based to/from lane-based. However, the issues, as source of potential inconsistencies, are:

• Vehicles view inter-model.

• Location of boundaries

4.1.1. Vehicles view inter-model

The process to move one vehicle from one model to the other requires the answer to a previous question: "Has the other model space to enter the vehicle?" If the answer is yes, then starts the process to transfer one vehicle

between models and according to the new behavioural model, then it calculates the new vehicle state (position and speed) and this information updates the internal structures to the upstream model. If the answer is no, the transfer of vehicle is not done and it creates the boundary conditions creating a fictitious vehicle stopped at the end of the lane.

In order to implement this mechanism of vehicle view inter-model, each submodel meso/micro takes from the whole network representation only the links defined as meso/micro (depending on the submodel) and creates an internal network representation. Figure 7 depicts a network partition where the internal representation of the mesoscopic network is formed by sections A, C and F (represented in grey) and the internal representation of the microscopic network is formed by sections B and E. However, for having information about the traffic state of the complementary model, it is necessary to add to each internal representation the next sections downstream. For instance the network in Figure 7, the internal representation of the mesoscopic model has to include section B tagged as external and the internal representation of the microscopic network has to include sections F and C tagged as external (see Figure 8).

Figure 7 Network partition with Meso (in grey) and Micro (in green) links

Figure 8 Internal representation of a) microscopic network and b) mesoscopic network

The process for updating the microscopic model could be summarized, taking into account that is a time dicrete simulation:

For each section s

For each Lane in section s

If section is tagged as external then

Update lane, reading last vehicle in mesoscopic model

Update all vehicles in lane, applying microscopic behavioural models endIf endFor endFor

And then the equivalent process for updating the mesoscopic model could be summarized, taking into account that is event oriented simulation:

For each event ev

If ev is Syncronization Event

For each section s tagged as external

Update lane, reading last vehicle in microscopic model endFor

Resolve Mesoscopic event endif endFor

4.1.2. Location of boundaries

The only limitation imposed by both models, considering the internal represetantion in each submodels and the update procedure explained above, is having all entrance legs in a node represented in the same submodel (in Figure 7 the entrance legs of the node located at the right are sections B and E represented as microscopic). This restriction is only imposed by the gap acceptance model applied in both models for deciding whether a vehicle can cross an intersection or not when there are conflicting movements. The gap acceptance models in micro and meso are slightly different due to the vehicle information available to each one (for instance the micro model considers acceleration, deceleration, meanwhile the mesoscopic model considers only the acceleration when a vehicle is stopped).

4.2. Target lanes calculation in meso-micro

One aspect to point out is the consistency between a meso or micro simulation compared with the hybrid simulation in terms of the lane changing models. The lane changing model implemented in Aimsun (2010) is based on a decision tree where one component is the decision of target lanes in each section. This decision is not only based on the traffic conditions present in the section but include the traffic conditions and the feasible lanes for reaching the turning movements determined in its path plan (a maximum distance exists that determines the look-ahead capability of each vehicle) and possible obstacles (Incident presence, Compulsory reserved lanes, Closed lanes, Presence of a Public Transport stop, in case of a Public transport vehicle).

Figure 9 depicts an example of network for calculating the target lanes. Assume that a vehicle is in section 271 and its path plan determines section 279 as final section, then the yellow line determines the sequence of target lanes calculated. However, if the vehicle is public transport vehicle the sequence of target lanes is determined by the blue line, because in section 275 there is a bus stop defined for this vehicle.

This approach is followed by each submodel, but the main drawback is each model has its own network representation without having information about the complimentary network of the other model. For example, in Figure 9, if sections 273, 274 and 275 are defined as micro and the rest as meso then the model that calculates the target lanes for a vehicle in section 271 (considered as meso) doesn't have information ("blind" model) about the

network defined downstream because there is a break of model. In that case the target lane model calculates all lanes in section 271 as feasible, because it only considers the next turning to section 273.

For solving this problem, the Aimsun hybrid model has a global look ahead procedure that it is independent of the submodel (meso/micro) in order of having a global view of the path plan of a vehicle.

Figure 9 Example of look a head 5. Computational Results

The computational results has been conducted complement the results obtained in (Casas at al., 2010) adding the analysis of adding the global look ahead.The computational results has been conducted in a part of the Madrid network, a mixed network with an urban definition complemented with part of the M-30, an urban highway. The main characteristics of the topology are: 327 centroids, 1375 intersections, 3591 sections with a total length of 570 Km (1160 Km as total length considering all lanes).

The traffic demand used contains the morning no-peak hour and the morning peak-hour, including the transition between both states, which means the demand from 6:00 am to 9:30 am, with a 15 minutes time slice, with a total demand of 409502 trips. The demand profile is an increasing demand until 8:00 am and then this demand is stabilized around 35000 vehicles every quarter of hour.

Taking into account the requirement of a set of intersections requires the detailed detection measurement for applying a future public transport pre-emption system, the network is partitioned into the mesoscopic network and the microscopic network. Figure 10 shows the partition of the network according to the behavioural model of the sections, where the sections displayed in green correspond to the mesoscopic model and the sections in blue correspond to the microscopic model (the total number of microscopic sections are 191, with 55 nodes considered as micro). Using this network, three different scenarios are compared:

• Scenario using only Mesoscopic simulation (Meso)

• Scenario using Hybrid (meso/micro) simulation without global look ahead

• Scenario using Hybrid (meso/micro) simulation witht global look ahead

Considering the validation of the real data set compared with the simulated outputs, the aggregated results of each scenario, globally all correct, could be summarized as:

Figure 10 The type of sections (meso/micro) of the Madrid network.

Table 1 Comparison of scenario Meso, Hybrid without global look ahead and Hybrid with global look ahead

MOE Meso Hybrid without global look ahead Hybrid with global look ahead

Vehicles gone out 365.159 345.396 362.483

Vehicles waiting out 20.318 36.935 20.418

Regression R2 0,90 0,86 0,90

Flow [veh/h] 103.628 104.349 102.868

Density [veh/km] 13,43 12,61 13,26

Speed [km/h] 39,82 39,56 39,82

Travel time [sec/km] 115,96 117,38 116,18

Delay time [sec/km] 54,81 61,54 55,33

The local analysis is performed using the GEH index, FHWA (2003) and Traffic Apparaisal Highways Agency (1966). GEH is an statistic used in traffic engineering to compare two sets of traffic data. In general two sets of traffic volumes. Although its mathematical form is similar to a chi-squared test, it is not a true statistical test. It is an empirical formula that has proven to be rather useful.It is defined as:

Where w and v are the simulated and observed counts respectively

2(w - v)2

However analyzing the local results using the GEH index for each section and plotting the GEH index (FIGURE 11 depicts the value of GEH of each link, where green is a good value with GEH <5, red is a regular value with a GEH >5 and <10 and blue represents point with a bad value), the Hybrid experiment without the global look ahead has difficulties to reproduce the traffic dynamics in two major roundabouts. The explanation of the incorrect GEH index in these two points is because the selection of the target lanes in roundabouts is crucial have the complete information of the path plan, selecting the correct entrance/exit lanes.

Running the same simulation using the Hybrid model with global look ahead feature, we get a better global results (see table 1) and comparing using the same GEH index, there is a local improvement because the discrepancies in these two roundabouts disappears ( see Figure 12)

Regarding to the computational time and taking as reference the microscopic simulation, the mesoscopic simulation of the same experiment takes a 10% of the microscopic simulation time, meanwhile the hybrid model this percentage increases slightly to 13%.

6. Conclusions

A main result of this study is the equivalence of the mesoscopic model and the hybrid model in terms of the validation of the real data set compared with the simulation outputs and the consistency of the traffic dynamics using a micro or meso model.

Once the consistency of the model is guaranteed, the critical point is the selection of the target lanes considering the look ahead capability of the vehicles. When there is not a global look ahead every time there is a break of model from meso to micro or micro to meso, the vehicles are losing information and the selection of the target lanes doesn't anticipate the downstream conditions. This assumption is corroborated with the computation results that depicts the main discrepancies were located in roundabouts. This problem is solved adding the global look ahead, independent of each submodel, improving the global and the local results.

Figure 10 GEH index by position

References

Aimsun, 2010, User Manual, TSS-Transport Simulation Systems, Barcelona, Spain.

V. Alexiadis, Role of Simulation in Corridor Management, presented at TRB2007 Simulation Workshop, Washington, January 2007.

BarcelóJ. and Casas J. (2006a), Stochastic heuristic dynamic assignment based on AIMSUN Microscopic traffic simulator, Transportation Research Records No. 1964, pp. 70-79.

Barceló J., Casas J., García D. and Perarnau J., (2006b), A hybrid simulation framework for advanced transportation analysis, Proceedings 11th IFAC Symposium on Control in Transportations, (IFAC-CTS2006) University of Delft

Barceló J., Casas J.,García D., and Perarnau J., (2005) Methodological notes on combining macro, meso and Micro Models for Transportation Analysis, Paper presented at the Workshop on Modeling and Simulation, Sedona.

Ben-Akiva M., Bierlaire M., Koutsopoulos H. and Mishalani R. (2002). Real-time simulation of traffic demand-supply interactions within DynaMIT, in M. Gendreau and P. Marcotte (eds), Transportation and network analysis: current trends. Miscellanea in honour of Michael Florian, Kluwer Academic Publishers, Boston/Dordrecht/London.

Casas J., Perarnau J., Torday A. (2010), Conceptual Approach Of A Hybrid Meso/Micro Simulation And Its Implementation For Traffic Signal Management Systems For Large Scale Networks. , Proceedings of the 89th TRB Annual Meeting.

Burghout W. (2004), Hybrid microscopic-mesoscopic traffic simulation, Doctoral Thesis, Royal Institute of Technology, Stockholm, Sweden.

Burghout W., Kotsopoulos H. and Andréasson I.(2005), Hybrid Mesoscopic-Microscopic Traffic Simulation, Proceedings of the 83rd TRB Annual Meeting.

Florian M.,Mahut M. and Tremblay N. (2001), A Hybrid Optimization-Mesoscopic Simulation Dynamic Traffic Assignment Model, Proceedings of the 2001 IEEE Intelligent Transport Systems Conference, Oakland, pp. 120-123.

FHWA (2003), Guidelines for applying Traffic Microsimulation Software.

Friesz T., Bernstein D., Smith T., Tobin R. and Wie B. (1993), A variational inequality formulation of the dynamic network user equilibrium problem. Operations Research, 41:179-191.

Janson B. (1991) Dynamic Assignment for Urban Road Networks, Transpn. Res. B, Vol. 25, Nos. 2/3, pp. 143-161.

Jayakrisham R., Mahmassani H. and Yu T., (1994) An Evaluation Tool for Advanced Traffic Information and Management Systems in Urban Networks. Transportation Research C, 2C (3), pp. 129-147.

Traffic Appraisal in Urban Areas, Highways Agency, Manual for Roads and Bridges, Volume 2, Section 2, Department for Transport, London, UK, 1966.