markovian modeling of urban traffic flows in coexistence with urban data streams

20
[email protected] SEC Markovian Modeling of Urban Traffic Flows in Coexistence With Urban Data Streams Vahid Moosavi Simulation platform, Future Cities Lab, ETHZ Supervisor: Professor Ludger Hovestadt Chair for Computer Aided Architectural Design, Department for Architecture, ETH Zürich 26 April 2013 1

Upload: vahid-moosavi

Post on 14-Apr-2017

75 views

Category:

Data & Analytics


0 download

TRANSCRIPT

Page 1: Markovian Modeling of Urban Traffic Flows in Coexistence With Urban Data Streams

[email protected]

Markovian Modeling of Urban Traffic Flows in Coexistence With Urban Data Streams

Vahid MoosaviSimulation platform, Future Cities Lab, ETHZ

Supervisor: Professor Ludger HovestadtChair for Computer Aided Architectural Design, Department for Architecture, ETH Zürich

26 April 2013

1

Page 2: Markovian Modeling of Urban Traffic Flows in Coexistence With Urban Data Streams

2

Multi-layer modeling and the curse of dimensionality…We take different layers (dimensions) and want to mimic the behavior. For example in Traffic modeling:• Shortest Path and rationality??!! • Traffic congestions?!• Traffic Lights?!!• Lots of other unknown elements that we don’t

know yet and in fact manipulate.

…Curse of Dimensionality…Complicated models, but not complex

Page 3: Markovian Modeling of Urban Traffic Flows in Coexistence With Urban Data Streams

3

Page 4: Markovian Modeling of Urban Traffic Flows in Coexistence With Urban Data Streams

4

Page 5: Markovian Modeling of Urban Traffic Flows in Coexistence With Urban Data Streams

5

Rational (Specific ) Models

Complex (Pre-specific ) Models

Properties of the system for modeling

Possible Relations (types and num

bers)

Multi-Agent Systems

Urban Cellular automata

Urban Dynamics

Basic Statistics(Hypothesis Testing)

Urban Metabolism

Natural (Deterministic)

Models

Urban ScalingSocial Physics

Fractal Models

Complexity and the Limits of Model-ability in Rational Way

It is not about more data or more computing power, we need an abstraction from

the concept of rational modeling.

Page 6: Markovian Modeling of Urban Traffic Flows in Coexistence With Urban Data Streams

6

An inversion in the concept of modeling

X YX Y

Model

Reality

Analysis

Synthesis Model

Reality

Celebration of ComputationCelebration of Connectedness

Celebration of AnalysisIf not then,

Page 7: Markovian Modeling of Urban Traffic Flows in Coexistence With Urban Data Streams

7

An inversion in the concept of modeling

X Y X Y

Celebration of Computation Celebration of Connectedness

Celebration of AnalysisIf not then,

Logic or rationale Or (descriptive

theories)

ObservationsObservations

Celebration of Computationsupports

Page 8: Markovian Modeling of Urban Traffic Flows in Coexistence With Urban Data Streams

8

An example From Language modeling…

Problems• Sentiment Analysis• Translation• Communication• …

Approaches for dealing with these problems1. Based on Grammar, Logic and

Model of the language. (Noam Chomsky)

2. Based on data-driven probabilistic models. (Originally by Markov and now in Google Translate)

… And maybe be a dialectical approach too...

On Chomsky and the Two Cultures of Statistical Learning: http://norvig.com/chomsky.html

Page 9: Markovian Modeling of Urban Traffic Flows in Coexistence With Urban Data Streams

9

Relational Model

Classic SpaceSyntax, London “The social logic of space,(1984)”

33,000+ taxicabs

GPS Trajectory of Taxicabs, Beijing, 2012

Inversion in Modeling

Rational Model

X Y X Y

Celebration of Computation Celebration of Connectedness

Page 10: Markovian Modeling of Urban Traffic Flows in Coexistence With Urban Data Streams

10

Video

Page 11: Markovian Modeling of Urban Traffic Flows in Coexistence With Urban Data Streams

11

An Experiment : Markovian Models in coexistence with data streams (using Taxi cabs GPS trajectories)

• Each Taxi produces a sequence of symbols. …It is telling its own story.

• Symbols could be road names, units of space, district names,…

• Sequence can be based on any time resolution. … we can construct a Markov Network encapsulating the transitions between states (symbols)

• Remark: The Markov network construction can be based on a specific time period (e.g. rush hours, weekends,…) or specific part of the city.

Possible functions• Simulation of traffic flow• Stationary distribution of cars• Road clustering• Road Engineering and scenario planning

– Finding critical roads– Road network sensitivity analysis– …

– As an opposing or complementary view to Chomsky, Linell presented interactionism: The sense-making ability of humans is rooted in social interaction; the mind is interactive, dialogical, social, shared, extended, distributed, etc.

Page 12: Markovian Modeling of Urban Traffic Flows in Coexistence With Urban Data Streams

12

Video : A sample Sequence

Page 13: Markovian Modeling of Urban Traffic Flows in Coexistence With Urban Data Streams

13

Experiment : Markovian Models in coexistence with data streams

0 0.5 0.5 0 0 0 0

0.5 0 0.5 0 0 0 0

0.450.45 0 0.1 0 0 0

0 0 0.5 0 0.5 0 0

0 0 0 0.1 0 0.45 0.45

0 0 0 0 0.5 0 0.5

0 0 0 0 0.5 0.5 0

CarID,Date,Lon,Lat,Symbol

100,2008-02-02 21:22:11,116.36263,39.93097,374100,2008-02-02 21:24:56,116.36708,39.92274,405100,2008-02-02 21:29:57,116.34696,39.92226,403100,2008-02-02 21:32:14,116.34557,39.91717,403100,2008-02-02 21:34:59,116.33843,39.92169,402100,2008-02-02 21:37:16,116.32875,39.92175,401100,2008-02-02 21:40:01,116.31468,39.9225,400100,2008-02-02 21:42:18,116.29511,39.92328,398100,2008-02-02 21:45:02,116.29542,39.9306,368

…,374,405,403,403,402,401,400,398,368,…

A sample stream of the data

A row stochastic Markov Matrix

1 2 3 4 5 6 7

1

2

3

4

5

6

7

Page 14: Markovian Modeling of Urban Traffic Flows in Coexistence With Urban Data Streams

14

Some Properties of Markov Chain in Urban road networkQuantity / Markov Network Trafic Network

Perron Eigenvector (dual) Vehicular density in the city network

Mean First Passage Times Average travel times for a pair of origin/destination

Kemeny constant Average travel time for a random trip

Perron Eigenvector (primal) Congested junctions in the network

Second Eigenvector (dual) Associates nodes to traffic sub-communities

1.Crisostomi, E., Kirkland, S., Shorten, R. (2011), A Google-like model of road network dynamics and its application to regulation and control. International Journal of Control

Page 15: Markovian Modeling of Urban Traffic Flows in Coexistence With Urban Data Streams

15

Future Steps

• Time series prediction for individuals• MCMC for multi-agent based simulation if needed : Data-Driven Simulation no

more direct theory or logic, but in principle we no longer need simulation but just analysis on top of data-driven models. For example, there is no need to be able mimicking the behavior of one day of a city, with urban data streams, we can watch it. We should go back to the history of simulation as a numerical approximation to Analytical models, which was the celebration of computing power, but now the issue is not about the computing power, it is about the limit of the thing (model based on theories) which are being computed. It is a limit of model-ability. Then, urban data streams brings a new capability for us.

Page 16: Markovian Modeling of Urban Traffic Flows in Coexistence With Urban Data Streams

16

• Markov Modeling of Singapore Ezlink Data• Based on important link in the Kemeny Analysis, run again the steady state probability

without that area.• Validation: Use power k of Markov and then compare with the result in K steps based on

empirical data• Predicting the future states by power of Markov Chain• Caclulating and visualizing the other network measures• Accessibility analysis using Mean first passage time: one measure can be just a an average

and deviation • Use SOM to compare different features such as Kemeny constant effect, First Eig, Average

Mean First Paassage time, Other features such closeness, betweenness, other network features

Page 17: Markovian Modeling of Urban Traffic Flows in Coexistence With Urban Data Streams

17

Results

Page 18: Markovian Modeling of Urban Traffic Flows in Coexistence With Urban Data Streams

18

Thanks!

Page 19: Markovian Modeling of Urban Traffic Flows in Coexistence With Urban Data Streams

Urban Data Streams Planning Interventions

Markov Chain (MC) Construction

Updating MC periodically

Urban Segments

Regional Scale

Transition Time

Selected Time Period Traffic Community Detection

Real Time Traffic Flow

Road network Engineering

Expected Empirical Travel Times

Network Analytics

City

Mining and AnalysisModeling

Page 20: Markovian Modeling of Urban Traffic Flows in Coexistence With Urban Data Streams

20