2016‐12‐15

1

Adaptive Ramp ControlThe Gardiner Expressway Case Study

Baher Abdulhai

Kasra RezaeeDepartment of Civil Engineering

Congestion SolutionsMore Supply

Less Demand

Intelligence for managing both supply and demand

2016‐12‐15

2

Control Approach Characteristics

1. “Pacing” Beats “Rushing”

2. Real Time: Closed Loop Optimal Control

3. Role for Artificial Intelligence

3

1. Pacing Beats Rushing

4

2016‐12‐15

3

Always measuring and mapping system state to optimal actions Better chance to be on top of changes But:

– How to get the optimal control policy

Possibly let the controller learn it: Machine Learning and AI

5

Traffic Network

Optimal Control

Law/Policy

Real Time Monitoring and Measurements

Optimal Control Actions

2. Closed Loop Optimal Control

3. Artificial Intelligence:Self-Learning the Optimal Control Law

2016‐12‐15

4

AI: Reinforcement Learning

Environment

Agent

State RewardAction

RL: Illustrative Demo

Baher

Source: https://www.youtube.com/watch?v=DCjbk4m1G6I

2016‐12‐15

5

RL: Another Illustrative Demo

Baher

Baher Abdulhai

Ramp Metering

Why Ramp Metering?1. Pacing demand: avoid congestion due to demand

exceeding capacity and resulting capacity breakdown

2. Avoid blockage of exit ramps3. Influence route choice behavior4. Enhance traffic safety:

• less congestion• safer merging

2016‐12‐15

6

Baher Abdulhai

Congestion Avoidance

Baher Abdulhai

Off Ramp Blockage

qin

d

qcap

)dq(1 cap

qcapd

γqin

qin

d

qcap

2016‐12‐15

7

Baher Abdulhai

Impacting Route Choice

Baher Abdulhai

Summary of Today’s Status Quo

Under-utilization of freeway corridors and networks due to recurrent and non-recurrent congestion:– high demand creates flow breakdown and congestion

causing loss of capacity

– downstream of congestion (empty stretch ahead)

– off-ramp blockage (stuck on the freeway)

– suboptimal utilization parallel arterial

– reduced safety

– increased pollution

2016‐12‐15

8

Baher Abdulhai

Ramp Metering Benefits in the Literature

30% more vehicles served during rush hours

Improved service for all users

Reduced urban network load

> 50% reduction of total time spent

Efficient response to incidents

Increased traffic safety

Decreased fuel consumption and environmentalpollution

Baher Abdulhai

Why NOT RM? Fallacy:

– RM benefits people on the freeway at the expense of those entering from the ramp

– RM causes traffic to use the surface street, i.e. benefit freeway at the expense of surface streets

Fact:– ramp queues do not mean dis-benefit to surface streets. As

the overall throughput of the freeway is improved, more surface street traffic can now use the freeway. i.e. benefit both.

– Coordinated RM does not penalize later entries (see Gardiner case)

2016‐12‐15

9

Baher Abdulhai

Ramp Metering Categories

Fixed Time Metering:– Mainly off line– Based on historical demand– Not responsive to real time traffic dynamics

Traffic Responsive:– Realtime– Regulator Approach:

– e.g. ALINEA

– Optimal Control Metering – e.g. RL

Control of Multiple On-ramps

18

2016‐12‐15

10

Ramp Metering the Gardiner

19

Gardiner Expressway westbound in Toronto from DVP to Humber Bay

Traffic from Don Valley Parkway

Humber Bay

Scenarios• Base Case• RLRM-I : Isolated • RLRM-IwQO : Isolated with queue override• RLRM-C : Coordianted• ALINEAwLC : ALINEA with linked control

BaseCase RLRM-IRLRM-IwQO

RLRM-CALINEAw

LCTTT 10276 5360 5665 5104 6660TTTml 6998 4147 4768 4249 4601

0

2000

4000

6000

8000

10000

12000

Tota

l tim

e sp

ent

on

net

wo

rk

(Veh

.hr)

48% 45% 50% 35%

Network Total Travel Time

20

2016‐12‐15

11

Downtown Origins Travel Times

21

14.34

6.34

9.32

6.70

16.65

6.19

8.81

6.57

16.27

5.47

8.75

6.75

12.13 11.79

9.478.68

BASECASE RLRM-I RLRM-IWQO RLRM-C

Travel time from different origins to west end of network

DVP Jarvis York Spadina

Local vs. Coordinated

22

14 16 18 20Time of day (hour)

p(

)

Base Case (Jameson Closed)

20

40

60

80

100

13 14 15 16 17 18 19 20 210

50

100

150

200

250

300

Time of day (hr)

Tot

al V

ehic

les

in Q

ueue

BaseCase (Jameson Closed)

13 14 15 16 17 18 19 20 210

0

0

0

0

0

0RLRM-I

Time of day (hr)

13 14 15 16 17 18 19 20 21

RLRM-IwQO

Time of day (hr)

YorkSpadina

Wes

tbou

nd D

irect

ion

13 14 15 16 17 18 19 20 210

0

0

0

0

0

0

Time of day (hr)

RLRM-C (coordinated)

Spadina

York

14 16 18 20Time of day (hour)

RLRM-I traffic speed

20

40

60

80

100

14 16 18 20Time of day (hour)

RLRM-IwQO traffic speed

20

40

60

80

100

14 16 18 20Time of day (hour)

RLRM-C traffic speed

20

40

60

80

100

2016‐12‐15

12

Conclusion

23

Least overall TTT

Reduced and

balanced queueing

Best overall travel times

from all ramps

Thank You!