NEXT GENERATION OF SIMULATION

1
NEXT GENERATION OF SIMULATION

• Jaume Barceló
• Department of Statistics and Operations Research
• Universitat Politècnica de Catalunya

2
KEY WORDS SINTHESIZING TRENDS AND CHALLENGES IN
TRAFFIC SIMULATION

• INTEGRATION (Of modeling approaches)
• NEW CORE MODELS
• Car-following, Lane Changing.
• DYNAMIC TRAFFIC ASSIGNMENT/DYNAMIC TRAFFIC
  EQUILIBRIUM
• IMPROVED MODELING APPROACHES
• Micro, Meso, Hybrid.
• MODEL CALIBRATION AND VALIDATION

3
INTEGRATION (Of modeling approaches)

• Advanced transport analysis must use various
  modeling approaches which must interact in an
  efficient way
• Efficient integration has requirements beyond the
  usual interfaces based on file data exchange
• Actual integration should be based on software
  engineering, exploiting common databases and data
  models, allowing a unique network representation
  shared by all modeling approaches.

4
NEW CORE MODELS Car-following, Lane Changing.

• The increasing use of microscopic simulation in
  more complex situations cannot always be
  efficiently addressed by current car-following,
  lane changing and similar core models.
• NGSIM research is developing new core models and
  a new philosophy
• A future simulation software with two main
  components
• A simulation engine and
• A wide library of core models which could be
  plug-in and plug-off by analysts according to
  simulation objectives or circumstances, i.e. from
  global to local models

5
IMPROVED MODELING APPROACHES Micro, Meso,
Hybrid.

• Meso capturing the high level dynamic traffic
  phenomena in large areas and
• Micro getting into the very details of the full
  dynamics in smaller areas.
• Meso and micro must work integrated and the best
  way is by means of hybrid meso-micro approaches

6
DYNAMIC TRAFFIC ASSIGNMENT/DYNAMIC TRAFFIC
EQUILIBRIUM

• DTA is a key concept for dynamic analysis.
• It is based on path calculations, estimation of
  dynamic path flow rates and network loading.
• An efficient way of implementing meso-micro
  integration while ensuring consistency is
  structuring DTA as a task independent of the
  modeling approach used. (i.e. a DTA Server)

7
MODEL CALIBRATION AND VALIDATION

• Model calibration and validation is currently a
  major bottleneck in traffic simulation practice.
• Tools to assist analysts in checking model
  consistency (static and dynamic) and designing
  and conducting the simulation experiments based
  on sound, proven techniques are necessary
• A promising technique successful the Simultaneous
  Perturbation Stochastic Approximation which has
  proven to be successful in other simulation
  domains is an example of the trends to follow.

8
A FEW WORDS ON INTEGRATION (Of modeling
approaches)
9
URBAN PLANNING?TRANSPORT PLANNING?OPERATIONAL
PLANNING
EMME/2 SATURN
GIS/CAD
AIMSUN
10
COMBINING STRATEGIC AND OPERATIONAL PLANNING

• IDENTIFIED TIME AGO AS AN EXTENSION TO STRATEGIC
  PLANNING
• WINDOWING INTO A POTENTIALLY CONFLICTING SUBAREA
  TO CONDUCT A DETAILED OPERATIONAL ANALYSIS
• PRACTICAL IMPLEMENTATIONS
• EMME/2 and AIMSUN
• SATURN and AIMSUN
• VISUM and VISSIM
• TRIPS and DYNASIM
• AIMSUN PLANNER AS A COMPONENT OF AIMSUN NG AND
  AIMSUN 5.1

11
SOME CRITICAL ASPECTS

• MACROSCOPIC MODELS (Usually based on a User
  Equilibrium approach) USE
• A NETWORK REPRESENTATION (I.E. LINK-NODE)
• A DEMAND MODELING (I.E. OD MATRIX, CENTROIDS,
  CONNECTORS)
• MICROSCOPIC MODELS USE
• A DETAILED GEOMETRY REPRESENTATION
• A DEMAND MODELING (I.E. OD MATRIX, CENTROIDS,
  CONNECTORS)
• CONSISTENCY ISSUES
• LINK/SECTIONS CORRESPONDENCES
• CENTROIDS AND CONNECTORS CORRESPONDENCES
• MODEL SYNCHRONIZATION
• HOW CHANGES IN ONE MODEL ARE TRANSLATED TO THE
  OTHER MODEL
• INEFFICIENCY OF FILE EXCHANGE INTERFACES

12
NEW TRENDS

• THE EVOLUTION OF ADVANCED TECHNOLOGIES AND THEIR
  APPLICATION TO MODERN TRAFFIC MANAGEMENT SYSTEMS
  DEMANDS A DYNAMIC VIEW
• BRING INTO THE PLANNING ARENA TIME DEPENDENT
  TRAFFIC PHENOMENA AS QUEUE SPILLBACKS OR THE TIME
  EVOLUTION OF CONGESTION
• MESOSCOPIC TRAFFIC SIMULATION MODELS ARE THE
  RESPONSE TO THESE REQUIREMENTS
• DYNASMART
• DYNAMIT
• DYNAMEQ
• MEZZO
• ..

13
WHATS BEST

• THE QUESTION IS NOT WHETHER ONE APPROACH IS
  BETTER OR MORE APPROPRIATE THAN OTHER
• OR IF THERE IS A UNIQUE APPROACH THAT CAN REPLACE
  SATISFACTORILY ALL OTHERS
• BUT WHICH IS THE MOST APPROPRIATE USE OF EACH
  APPROACH AND HOW CAN THEN WORK TOGETHER IN A
  COMMON FRAMEWORK

14
FROM MACRO TO MESO AND MICRO

• GENERALIZE THE FORMER MACRO-MICRO INTERFACES TO A
  COMBINATION OF MACRO, MESO AND MICRO
• BY MEANS OF AN INTEGRATED SOFWARE ENVIRONMENT TO
  SUPPORT A TRANSPORT ANALYSIS METHODOLOGY THAT
  COMBINES THE THREE APPROACHES, OVERCOMING
  CONSITENCY AND SYNCHRONIZATION PROBLEMS

15
Three Types of Models Used in Corridor Analysis
(V. Alexiadis, TRB 2007)
16
Mesoscopic Modeling (V. Alexiadis, TRB 2007
17
An Integrated Analysis Methodology(V. Alexiadis,
TRB 2007)

• Revised Trip Tables
• Refined Travel Times

Enhanced Performance Measures
No
Meso- and/or Micro-simulation

• VMT/VHT/PMT/PHT
• Travel Time/Queues Throughput/Delay
• Environment
• Safety

Dynamic Assignment
Convergence ?
Yes
Pivot Point Mode Choice
Refined Transit Travel Times

• Refined Trip Table(Smaller Zones and Time
  Slices)
• Refined Network

Benefit Valuation
Outputs

• Benefit/Cost Analysis
• Sensitivity Analysis
• Ranking of ICM Alternatives

Cost of ImplementingStrategies
User Selection of Strategies
18
A METHODOLOGICAL PROPOSAL

• FROM A MODELING POINT OF VIEW THERE ARE TWO MAIN
  ISSUES TO ACHIEVE THIS OBJECTIVE
• COMPUTER MODELING ISSUES
• AN EFFICIENT INTEGRATION OF TRANSPORT ANALYSIS
  TOOLS REQUIRES AN SPECIFIC COMPUTER MODELING
  APPROACH TO OVERCOME THE INEFFICIENCIES AND
  LIMITATIONS OF TRANSPORT MODELS BASED ON
  DIFFERENT NETWORK REPRESENTATIONS COMMUNICATING
  BY MEANS OF FILE EXCHANGES
• TRAFFIC MODELING ISSUES
• CONCERNING THE MODELING PARADIGMS TO BE USED AT
  EACH LEVEL AND THEIR COMPATIBILITY.

19
CONCEPTUAL ARCHITECTURE OF THE INTEGRATED
ENVIRONMENT AND THE EXTENSIBLE OBJECT MODEL
20
AIMSUN NG (CONCEPTUAL ARCHITECTURE)
An integrated
enviroment for transport analysis
21
MULTILEVEL NETWORK REPRESENTATIONNode-Link
(left) Detailed Geometry (right)
22
SIMULTANEOUS NETWORK ANALYSIS User Assignment
(Left), Mesoscopic Simulation (Center),
Microscopic Simulation (Right)
23
CONCEPTUAL DIAGRAM OF THE METHODOLOGICAL PROCESS
COMBINING MACRO?MESO?MICRO
MACRO LEVEL TRANSPORT PLANNING MODEL OF A
REGIONAL OR METROPOLITAN AREA
24
COMMENTS ON NEW CORE MODEL DEVELOPMENTS(NGSIM
CONTRIBUTIONS)- Car following - Lane
Changing? Made possible by the availability of
new more accurate traffic data NGSIM data
25
Problem Statement Oversaturated Freeway Flow
(UC Berkeley)

• Existing Simulation Approaches
• Do not accurately model oversaturated traffic
  conditions such as repeated stops and starts,
  increased lane changes to position onto perceived
  moving lanes, or in the presence of tall
  vehicles, and large vehicle headways
• Use additional rules and parameters to the basic
  desired headway and gap-acceptance based
  car-following and lane changing models
• Introduce a large number of parameters that
  generally cannot be readily observed in the field

26
Car-following Model (UCB)
Jam Spacing
Wave travel time
Newells simplified CF model
Maximum Acc
Free Flow speed
Safety Constraint
Maximum Dec
27
Lane Changing Mechanism (UCB 1)
Car-following rule for Lane changing pending
xMIN( xCF(Leader) , xCF(Veh Down) )
x xCF(Leader)
Try to pass and find next gap
downstream vehicle
upstream vehicle
Veh down
Veh up
Leader
LC
Conflict point
Request Cooperation
28
Lane Changing Mechanism (UCB 2)
Car-following rule for cooperating vehicle
xMIN( xCF(Leader) , xCF(LC) )
End cooperation
upstream vehicle
Leader
Veh up
Coop
LC
Conflict point
Request Cooperation
29
Lane Changing conflict (UCB 3)
If there exist conflict in lane changing
xMIN( xCF(Leader) , xCF(Veh Down) )
LC1 will yield to LC2
else x xCF(Leader)
LC2 will pass LC1
downstream vehicle
upstream vehicle
LC
Veh up
LC2
Leader
LC1
Request Cooperation
30
(No Transcript)
31
Wave Travel Time (UCB)

• t time between the action of leader vehicle
  and the following vehicle

32
Wave Speed (UCB)

• Wave speed sjam / t (?gjam) / t
• US101
• Mean18.07km/hr,11.22miles/hr
• I-80
• Mean19.59km/hr,12.18miles/hr

33
Max Acceleration Deceleration (UCB)

• Extracted from NGSIM trajectories data US101
• Passenger cars (n4163)
• Mean4.516(m/s2), std 0.808
• mean -4.398(m/s2), std0.827

34
MIT LANE SELECTION MODEL
35
MIT NGSIM LANE SELECTION MODEL Discrete Choice
Random Utility with 31 parameters Implemented in
AIMSUN using its Extensible Object Model
36
DYNAMIC TRAFFIC ASSIGNMENT/DYNAMIC TRAFFIC
EQUILIBRIUM
37
CONCEPTUAL ALGORITHMIC FRAMEWORK FOR DYNAMIC
TRAFFIC ASSIGNMENT

• Principles
• Components of DTA models
• Method for determining path-dependent flow rates
  on the paths in the network
• Dynamic equilibrium algorithm
• Route Choice stochastic model
• Dynamic network loading method
• Simulation model
• Mesoscopic
• Macroscopic
• Analytical model
• Preventive
• Combines the experienced travel times with
  conjectures to forecast the temporal variations
  in flows and travel costs.
• Reactive
• Evolution of flows when users make route choice
  decisions based on experienced travel times

38
PREVENTIVE DTA IN AIMSUN

• Link Cost Function

input cost of link j at iteration n at time
interval t
ouput cost of link j at iteration n at time
interval t
input cost of link j at iteration n1 at time
interval t
EXPECTED LINK COST
EXPERIENCED LINK COST
39
THE LINK COST CALCULATION (Prediction based on
exponential smoothing from previous iterations)
40
PATH-DEPENDENT FLOW RATES BASED ON DYNAMIC USER
EQUILIBRIUM MODELS
41
THE AD HOC MSA ALGORITHM
42
THE MESO APPROACH NETWORK LOADING (I)

• Vehicle based event scheduling approach
• vehicles generation
• entrance of a vehicle into a link
• the transfer of a vehicle from one link to the
  next according to turning movements at
  intersections
• Approximate description of vehicles trajectories
  in the links
• The approach models the link dynamics splitting
  the link in two parts
• the running part that part of the link where
  vehicles are not yet delayed by the queue
  spillback at the downstream node account for
  different flow regimes and transition phases
• the queue part nodes are modeled according to a
  queue server approach (i.e. adapted from a GI/G/n
  model)

43
THE MESO APPROACH NETWORK LOADING (II)

• Individual vehicle dynamics in the running part
  approximated by a simplified car following model
  compatible with non linear macroscopic
  speed-density relationships on the link
• Accounting for different flow regimes
• And transition phases according to the three
  phase flow model

44
COMMENTS ON HYBRID MESO-MICRO MODELS
45
HYBRID MESO/MICRO SIMULATION
46
AUTOMATIC CREATION OF THE LOCALGATE-IN /
GATE-OUT DUMMY CENTROIDS
47
DTA SERVER

• Dynamic Traffic Assignment Server

48
GRAPHIC DEFINITION OF THE SUBAREA, AND NEW
CENTROIDS CONFIGURATION
49
MESO/MICRO LINK TRAVEL TIME CONSISTENCY
50
MESO/MICRO PATH TRAVEL TIME CONSISTENCY
51
INTEGRATION REVISITED FROM A SOFTWARE
ARCHITECTURE PERSPECTIVE
52
ENHANCED SOFTWARE ARCHITECTURE TO INTEGRATE MODELS
53
An Example of User Exploitation of the
ArchitectureUCB Testing of the Oversaturated
Freeway Algorithm

• AIMSUN SDK
• Replacing AIMSUN car-following and lane changing
  model
• Programming language C

A2Vehicle
A2VehicleBehavioralModel
A2VehicleBehavioralModelTest
A2VehicleTest
A2VehicleModelTestCreator
dll entry point for new model
Car-following and lane changing Logic
Vehicle behaviors
54
UCB Model Testing I-80

• Simulation 230PM-300PM
• Detector 7

55
CONCLUSIONS

• The integrated software environment presented in
  this lecture provides an efficient computer
  modeling solution to the problem of consistent
  network representation shared by various traffic
  modeling approaches based on different modeling
  paradigms as the macro, meso and microscopic
  approaches.
• This computer modeling solution enables a
  transport analysis methodology based on the
  exchange of information between the various
  modeling levels.
• It also provides a consistent basis for a common
  calculation of shortest paths based on the
  concept of Dynamic Traffic Assignment Server that
  makes meso and micro approaches compatible.
• The consistency of such compatibility has been
  proved.

56
SOME REFERENCES

• C. Chudhury, T. Toledo, M. Ben-Akiva (2004), Lane
  Selection Model, draft final report,
  Massachusetts Institute of Technology.
• TSS-Transport Simulation Systems (2005), AIMSUN
  Testing of the NGSIM Lane Selection Model, report
  presented at the Modeling and Simulation
  Workshop,85th TRB Annual Meeting.
• J. Barceló and J. Casas, (2006) Stochastic
  heuristic dynamic assignment based on AIMSUN
  Microscopic traffic simulator, Paper 06-3107
  presented at 85th Transportation Research Board
  2006 Annual Meeting, published in Transportation
  Research Records 1984, Fisher C.
• J. Barceló, J. Casas, D. García and J. Perarnau
  (2006), A hybrid simulation framework for
  advanced transportation analysis, Proceedings of
  the 11th IFAC Symposium on Control in
  Transportation Systems (IFAC-CTS2006), University
  of Delft.
• J. Barceló, J. Casas, D. García and J. Perarnau
  (2006), A methodological approach combining
  macro, meso and micro models for transportation
  analysis, Proceedings of the 13th World
  Conference on ITS, London (UK)
• H. Yeo, A. Skabardonis, J. Laval (2006), TO 9
  Oversaturated Freway Flow Algorithm, Interim
  Report, University of California, Berkeley