proactive vehicle re-routing strategies for congestion avoidance

19
Proactive vehicle re-routing strategies for congestion avoidance Juan (Susan) Pan * , Mohammad A. Khan * , Iulian Sandu Popa + , Karine Zeitouni + and Cristian Borcea * * New Jersey Institute of Technology, USA + University of Versailles Saint-Quentin-en- Yvelines, France DCOSS 2012

Upload: allie

Post on 23-Feb-2016

42 views

Category:

Documents


1 download

DESCRIPTION

Juan (Susan) Pan * , Mohammad A. Khan * , Iulian Sandu Popa + , Karine Zeitouni + and Cristian Borcea * * New Jersey Institute of Technology, USA + University of Versailles Saint-Quentin-en-Yvelines, France DCOSS 2012 . Proactive vehicle re-routing strategies for congestion avoidance. - PowerPoint PPT Presentation

TRANSCRIPT

Page 1: Proactive vehicle re-routing strategies for congestion avoidance

Proactive vehicle re-routing strategies for congestion

avoidanceJuan (Susan) Pan*, Mohammad A. Khan*, Iulian

Sandu Popa+, Karine Zeitouni+ and Cristian Borcea* *New Jersey Institute of Technology, USA

+University of Versailles Saint-Quentin-en-Yvelines, France

DCOSS 2012

Page 2: Proactive vehicle re-routing strategies for congestion avoidance

2

Traffic congestion: an ever-increasing problem

In 2010, congestion caused urban Americans a cost of $101 billions

By 2015, this cost will rise to $133 billion and the amount of wasted fuel will jump to 2.5 billion gallons

Increase road capacity? Optimize traffic signal control? Provide traffic guidance to drivers?

Page 3: Proactive vehicle re-routing strategies for congestion avoidance

3

Congestion avoidance using mobile sensing and actuation (1)

Smart phones (mobile sensors) & road-side sensors monitor traffic at fine granularity Mobile sensors can be vehicular embedded systems Road-side sensors: loop detectors, cameras, etc Demonstrated by other researchers

Traffic management service (TMS) collects data and estimates congestion in real-time

Page 4: Proactive vehicle re-routing strategies for congestion avoidance

4

Congestion avoidance using mobile sensing and actuation (2)

TMS provides real-time, proactive, individually-tailored re-routing guidance to drivers to prevent congestion Drivers provide their origin-destination

information Guidance is pushed to drivers’ smart phones

when signs of congestion are observed on their current route

Drivers may or may not follow the guidance The main focus of our research

Page 5: Proactive vehicle re-routing strategies for congestion avoidance

5

Comparison with existing work (1)

Google, Microsoft & Inrix: real time traffic info to compute traffic-aware shortest routes Reactively provide same

guidance for all drivers Problem: move congestion from

one spot to another▪ Similar to route oscillation in

computer networks

Page 6: Proactive vehicle re-routing strategies for congestion avoidance

6

Comparison with existing work (2)

Transportation researchers proposed dynamic traffic assignment models Iteratively calculate shortest paths and assign

routes to each driver to achieve the user/system equilibrium

Example of systems: DynaMIT, Contram Problems ▪ Tractability at scale (providing real-time guidance)▪ Ability to work when not all drivers are part of the

system▪ Robustness to drivers who ignore the guidance

Page 7: Proactive vehicle re-routing strategies for congestion avoidance

7

Outline

Motivation & related work Our 3 proactive re-routing strategies Simulation results Conclusion and future work

Page 8: Proactive vehicle re-routing strategies for congestion avoidance

8

The 4 phases of re-routing Road network represented as graph, with estimated

travel time as edge weight1. Travel time estimation

Greenshield’s model for travel time estimation

2. Traffic congestion estimation Density greater than threshold (δ=0.7)

3. Selection of candidate vehicles for re-routing4. Re-routing: alternative route computation and

assignment to drivers

Page 9: Proactive vehicle re-routing strategies for congestion avoidance

9

Selection of candidate vehicles for re-routing

Step 1: Detect road segments with signs of congestion

Blue: 1st level incoming segmentsGreen: 2nd level incoming segments

Step 2: Recursively select incoming segments to “congested” segment until depth L

Step3: Select vehicles on these road segments

Page 10: Proactive vehicle re-routing strategies for congestion avoidance

10

Our 3 re-routing strategies

Dijkstra’s Shortest Path (DSP) Computes one single shortest path for each driver Potential to switch congestion from one spot to

another Random K Shortest Paths (RkSP)

Compute k shortest paths for each driver and randomly pick one

Solves DSP problem, but could be far from optimal Entropy Balanced K Shortest Paths (EBkSP)

Prioritize candidate vehicles Compute k shortest paths for each driver and pick

the one with least popularity Improves on RkSP by choosing better paths

Page 11: Proactive vehicle re-routing strategies for congestion avoidance

11

EBkSP popularity entropy

Let (p1,…, pk) be the set of k paths that will be assign next

Let (r1,…, rn) be the union of all segments of (p1,…, pk), and (fc1,…,fcn) be the set of weighted footprint counters

Def: Entropy(pj) is the entropy of pj and is computed as

Def:

Def: the weighted footprint counter fci of a road segment i is: fci =ni х wi ni is the number of vehicles that are assigned to paths that include this segment, and wi is the weight of the road segment

Page 12: Proactive vehicle re-routing strategies for congestion avoidance

12

EBkSP example

a b c d e

f g h i j

k

V2, assigned path (fg, gh, hi, ij )V1, assigned path (ab, bg,gh, hi, ij )V3, assigned path (ch, hk )P1 (ab, bg, gh, hi, ij) fc(P1)(1,1,2,2,2) entropy=1.49P2 (ab, bc, ch, hi, ij) fc(P2)(1,0,1,2,2) entropy=1.16P3 (ab, bc, cd, di, ij) fc(P3)(1,0,0,0,2) entropy=0.58

?

Page 13: Proactive vehicle re-routing strategies for congestion avoidance

13

Evaluation Goals

Effectiveness average travel time Users’ experience number of re-routings Robustness average travel time as function of driver

compliance rate & system penetration rate Real-time CPU time

Simulation setup SUMO-0.13 microscopic simulator, open source, car

following model TRACI python library send commands to SUMO to assign

new paths

Brooklyn (1000 vehicles)Newark (908 vehicles)

Page 14: Proactive vehicle re-routing strategies for congestion avoidance

14

Average travel time

All strategies improve average travel time significantly EBkSP has the best performance

Compared to “no-reroute” reduces the travel time by up to 81% and 104%

Compared to DSP, it is better by 15% and 25%

Brooklyn(L=3,δ=0.7)

Brooklyn(L=4,δ=0.7)

13001500170019002100230025002700 noreroute DSP RkSP EBkSP

Aver

age

trav

el

time

(sec

onds

)

Newark(L=3,δ=0.9)

Newark(L=4,δ=0.9)

1500

2000

2500

3000

3500

4000

Aver

agte

tra

vel

time

(sec

onds

)

Page 15: Proactive vehicle re-routing strategies for congestion avoidance

15

Average number of re-routings

EBkSP has least number of re-routings Less distraction to drivers

Newark(L=3,δ=0.9)

Newark(L=4,δ=0.9)

1.71.92.12.32.52.72.93.13.33.5 DSP RkSP EBkSP

Aver

age

num

ber

of

rero

utin

gs

Brooklyn(L=3,δ=0.7)

Brooklyn(L=4,δ=0.7)

1.31.51.71.92.12.32.5 DSP RkSP EBkSP

Aver

age

num

ber

of r

erou

ting

s

Page 16: Proactive vehicle re-routing strategies for congestion avoidance

16

Average travel time vs. compliance and penetration rate

130014001500160017001800

DSP RkSP EBkSP

Compliance rate

Aver

age

trav

el

time

(sec

onds

)

0.2 0.313001500170019002100230025002700 noreroute DSP RkSP EBkSP

Compliance rateAv

erag

e tr

avel

tim

e(se

cond

s)

Performance increases with compliance rate up to a point Low compliance still much better than “no-reroute”

1400160018002000220024002600

DSPRkSPEBkSPnoreroute

Penetration rate (no road-side sensors)

Aver

age

trav

el

tim

e(se

cond

s)

1400160018002000220024002600

DSPRkSPEBkSPnoreroute

Penetration rate(with road-side senors)

Aver

age

trav

el

time(

seco

nds) Performance increases with penetration rate

With road-side sensors help to detect vehicular density, low penetration rate still much better than “no-reroute”

Page 17: Proactive vehicle re-routing strategies for congestion avoidance

17

CPU time

DSP lowest CPU time Dijkstra lower complexity than k shortest path O(E + V log(V )) vs. O(kV (E+V log(V )))

EBkSP better than RkSP Fewer origin-destination pairs due to better re-routing

Brooklyn(L=3,δ=0.7)

Brooklyn(L=4,δ=0.7)

01020304050607080

DSP RkSP EBkSP

Cpu

tim

e(se

cond

s)

Brooklyn(L=3,δ=0.7)

Brooklyn(L=4,δ=0.7)

1000

1100

1200

1300

1400

1500 DSP RkSP EBkSP

Tota

l num

ber

of O

D pa

irs

Page 18: Proactive vehicle re-routing strategies for congestion avoidance

18

Conclusion & future work

All re-routing strategies decrease significantly the average travel time EBkSP is the best—careful path selection ▪ 104% improvement compared to the “no rerouting” baseline▪ Lowers with 34% the number of re-routings

Good improvements are observed even for relatively low penetration/compliance rates

To improve scalability and real-time response, we plan to work on hybrid system architectures Offload part of computation to mobile nodes Use ad hoc communication in addition to Internet

communication

Page 19: Proactive vehicle re-routing strategies for congestion avoidance

19

Thank you !Acknowledgment: NSF Grant CNS-0831753

http://www.njit.edu/~borcea/invent/