Title: Dynamic Traffic Assignment V: Realtime Systems, Consistency in Predictive Information
1Dynamic Traffic Assignment VReal-time Systems,
Consistency in Predictive Information
Haris N. Koutsopoulos Northeastern
University M.I.T. Summer Professional Program
1.10s Modeling and Simulation for Dynamic
Transportation Management Systems July/August
2003
2Outline
- Introduction
- Real-time Information Generation
- DynaMIT
- Case Studies
- Conclusion
3Introduction
- Information media
- TV - Radio
- Internet
- Variable Message Signs
- Cellular phones
- On-board equipment
- Information mode
- Prescriptive
- Descriptive
4ATIS Internet
Houston http//traffic.tamu.edu/traffic.html
5ATIS Internet
Houston http//traffic.tamu.edu/traffic.html
6ATIS Internet
Paris http//www.sytadin.tm.fr
7Real-time Guidance and Traffic Information
Generation
- Information Attributes
- Reliable
- Unbiased
- Timely
8Real-time Information Generation
Broadcasting weather forecasts does NOT change
the weather
9Solution
- Traffic prediction incorporating
- Future OD flows
- Driver response to information
- Consistent information
- Result
- Prevent over-reaction
- Maintain system credibility
10Consistency
Path Splitsp
N
B
Informationi
Link Impedancet
A
11Consistency
- Supply Mapping
- Path splits Link Impedance
- Dynamic Network Loading
- Guidance Mapping
- Link Impedance Information
- ATIS
- Behavior Mapping
- Information Path Splits
- Behavior Models
12DynaMIT
Simulation-based
Dynamic traffic assignment system
With traffic prediction capabilities
- Generating consistent travel information
13DynaMIT
14DynaMIT
- Real-time applications DynaMIT-R
- Short-term planning applications DynaMIT-P
- Functionalities
- Dynamic Traffic Assignment
- Estimation and prediction of OD flows
- Estimation of current traffic conditions
- Prediction of future traffic conditions
- Incorporation of the effect of traffic
information - Fusion of historical information and real-time
sensor data - Rolling horizon mode
- Day-to-day learning and equilibrium models
15Supporting Utilities
- Calibration of model parameters
- Output processing and analysis
- Graphical User Interfaces
- Output visualization
- Network generation and editing
16Dynamic Traffic Assignment Framework
- Demand
- Origin-destination flows
- Micro simulator of travel choices
- Supply
- Mesoscopic Traffic simulator
- Demand-Supply Interactions
ATIS ATMS
Demand Simulator
Supply Simulator
State Estimation And Prediction
17Demand Simulator
- Key Elements
- Origin/Destination estimation and prediction
- Disaggregation-Aggregation process
- Travel behavior and response to information
through disaggregate behavioral models - Departure time
- Route choice
18Inputs
Inputs
List of drivers
Control
Guidance
Incidents
Supply Simulator
Mesoscopic
Simulator
En-route
Speed density realtionships Individual vehicles
Demand
Simulation
Congestion
Queues
Spillbacks
Network Conditions
Flows
Queues
Travel Times
Speeds
Densities
Assignment Matrix
etc.
19DynaMIT Overall Framework
20Demand-Supply Interactions
- Fixed point problem
- Find x f(x) x
- Algorithms
- Variations of
- xk1 g(xk, f(xk))
21DynaMIT State Estimation
22OD Estimation
OD Estimator
- x estimated OD flows
- xH historical (seed) OD flows
- A assignment matrix
- Ax estimated counts
- y observed sensor counts
- F1, F2 metric
Assignment Matrix A Maps OD flows to sensor
counts. Proportion of demand in period i for a
particular OD contributing to counts in period j
23Assignment Matrix
- Depends on
- Path travel times
- Path choice probabilities
Traffic
Historical
Counts
Database
- Congestion
- Unknown ? iterative solution
OD Estimation
OD Flows
24Assignment Matrix
25DynaMIT Predictive Information Generation
26(No Transcript)
27OD Prediction
- Predict O-D flows for intervals (h1,h2,)
- Deviations, Devh1
- Preserve historical information and structure
- Autoregressive process
- Devh1 ?0Devh ?1Devh-1 ?2Devh-2.
- Prediction
- ODh1Adjusted_Historical_ODh1 Devh1
28Rolling Horizon Operation
29System Design
30Applications
- Generation and provision of information to
travelers - Generation and evaluation of incident response
strategies/diversions - Provision of information to TMC
- Generation of emergency response
31Case Studies
- Irvine, CA
- Central Artery, Boston
- New York
- Switzerland
- Los Angeles
- Hampton Road, Virginia
32Case Study The Irvine Network
- Network
- Three major freeways
- Dense arterial network
- 298 nodes, 618 links, 1372 segments, 655 OD pairs
- Ramp metering
- Actuated signal control
- Data
- Static planning OD matrices
- 5 days of counts, speeds from 68 sensors
- 30-second data on freeway links
- 5-minute data on arterial links
- Scope
- Evaluation of estimation and predictive
capabilities
33Primary OD Pairs
34Evaluation Steps
- Calibration
- 3 days of data
- Validation
- Remaining 2 days of data
35Sensor Data Variability
36Calibration
- Supply Module
- Demand Module
37Supply Calibration
- Speed-density relationships
- Lane group capacities
- Intersection capacities
- Highway Capacity Manual
- Approximation of actuated strategies
38Demand Calibration
- Route choice model parameters
- Travel time
- Freeway bias
- O-D Estimation and Prediction
39Calibration Framework
- Sources of Errors
- OD matrix
- Route choice
- Traffic dynamics
- Aggregate data
- Flows, speeds
- Objective Function
- Measure of the difference between observed and
simulated values
40Calibration Iterative Process
41Path Choice Set
- Generation of reasonable paths for each OD pair
- Criteria
- Shortest paths
- Connectivity
- Random perturbation of travel times
42Calibration Results
- 15 min estimation interval
- Speeds (mph)
43Calibration results, contd
- 15 minute estimation interval
- Counts (veh/15 min)
44Validation
- Same sensor locations, different days (2 days)
- 15 minute estimation, 30 minutes prediction
- 5 minutes estimation, 60 minutes prediction
45Validation Results
- 15 minute estimation
- 30 minutes prediction
46State Estimation Results
- 15 minute estimation interval
- Counts (veh/15 min)
47State Estimation Results
- 15 minute estimation interval
- Counts (veh/15 min), freeway sensor location
48State Estimation Results
- 15 minute estimation interval
- Counts (veh/15 min), ramp sensor location
49State Estimation Results
- 15 minute estimation interval
- Counts (veh/15 min), arterial sensor location
50Prediction Results
- OD flows divided into classes
- High, medium and low
- Autoregressive process of degree 4 for each
class
- Results
- Estimation from 715 to 730 am
- Prediction from 730 to 800 am
51Prediction Results
- 15 minute estimation interval, counts (veh/15 min)
52Validation Results
- 5 minute estimation
- 60 minutes prediction
53Estimation Results
- 5 minute estimation interval, counts (veh/5 min)
54Prediction Results
- Estimation from 725 to 730 am
- Prediction from 730 to 830 am
- Deviations Devh
- Autoregressive process of degree 4
- Devh1 ?0Devh ?1Devh-1 ?2Devh-2.
- Prediction
- ODh1Adjusted_Historical_ODh1 Devh1
55Prediction Results
- 5 minute estimation interval, counts (veh/5
min) - estimation 725-730, prediction 730 to 830
56Impact of Information Results
- 715-815
- Incident
- 725 to 745
- 45 capacity reduction
57Impact of Information Results
- Sample of alternative paths
- Path Choice Set Generation
- Shortest paths
- Link elimination
- Random perturbation
58Impact of Information Results
- With predictive information
- Main O-D pair
- 18 reduction in travel time
- All O-D pairs with paths using the incident link
- 17 reduction in travel time
59Impact of Information Results
Without information
With predictive information
60Case Study Central Artery
- Impact of predictive information
- informed drivers
61Impact of Predictive Guidance
without guidance
with guidance
62Sensitivity Analysis
- Prediction Horizon 20 min
63Case Study TRANSMIT
- Network of detectors (EZ-PASS)
- Lower Westchester, NY
- Scope
- Diversion strategies for an incident management
system
- Network
- 1500 links
- 850 nodes
- 24 TRANSMIT sensors
- High market penetration
- High demand
64TRANSMIT Data
65Advantages of TRANSMIT Data
- Better observability of the network ? state
estimation - Direct observations
- Sub-path flows
- Point to point travel times (? reliable speeds)
- Route choice fractions
- Potential for lower measurement errors
66Role of DynaMIT
- Diversion strategies
- Generation
- Evaluation
- Refinement
- Guidance
- Generation of messages
- Dissemination to drivers
67Approach
- State estimation and prediction
- Incorporate additional information
- Traffic data fusion
- Loop detectors
- TRANSMIT sensors
68Impact of Additional Information
69Impact of Additional Information
70Case study Switzerland
- Large intercity network
- 1661 links
- 2177 segments
- 912 nodes
- 206 sensors
- 11 088 OD pairs
- 91 226 paths
- Free-flow TT 1.5h-2h
71Scope
- Pre-defined strategies
- Real-time decision-aid tools
72Los Angeles
- Support an incident management system
- Evaluate DynaMIT with different prediction,
estimation, and rolling horizon intervals - Use DynaMIT for
- Base case conditions
- Incident impacts
73Hampton Road, Virginia
- Calibration
- Off-line validation
- Off-peak
- Peak
- Incident
- On-line validation
74Conclusion
- Methodological issues
- Process real-time data
- Importance of behavior
- Solutions (DynaMIT)
- Estimation
- Prediction
- Predictive, consistent guidance
- Case Studies
- Validation
- Diversion strategies
- Real time
- Off-line