" Simulation based investigation of vehicle-to- vehicle dynamics and congestion“

(1)

of 28

DTU Management Engineering, Technical University of Denmark

" Simulation based investigation of vehicle-to- vehicle dynamics and congestion“

Andrea Vanesa Papu Carrone

Jeppe Rich

(2)

Agenda

 Introduction

 Research questions

 Theoretical background

 Experiment setup

 Simulation framework

 Scenario analysis: Results

 How / What should we plan?

 Conclusions, limitations and future research

(3)

of 28

 Improving infrastructure operations through intelligent traffic management systems

 For 2050: 7 – 61% trips performed in AVs

(Milakis et al.,2016)

Introduction (II)

INTRODUCTION / THEORETICAL BACKGROUND / EXPERIMENT SETUP / SIMULATION / SCENARIO 3 ANALYSIS / PLAN / CONCLUSIONS

VEHICLE BASED INNOVATION VEHICLE BASED INNOVATION

Autonomous Vehicles (AVs)

Mixed traffic environment (regular cars + AVs)

(4)

Consequences on the traffic environment:

 Network performance

 AVs demand and car ownership

 Mode split

 Traffic Safety

 Route planning

Potential impacts of introducing AVs

in congested traffic situations on motorways

FOCUS: MICROSCOPIC BEHAVIOR OF TRAFFIC

 Will AVs penetration to the market improve capacity and traffic flow?

 Is this improvement proportional to the % AVs? or will AVs smooth traffic flow even at low rates of

penetration?

Investigate the vehicle dynamics using simulation tools

(5)

of 28

5

Ongoing research

“The result shows that even 10% of vehicles driving with this strategy

significantly

reduce the duration and size of the congestion and, as a consequence, the

travel times

and the overall fuel consumptions. When running the simulation with a penetration

rate of 20% equipped vehicles, we observe only a short and insignificant congestion for the given traffic demand.”

“There is great potential for substantial improvements in network performance,

particularly in high-speed, high flow situations. However, there is evidence that at low penetrations, vehicles are not

able to make use of their enhanced capability. This work suggests that the

required proportion of Av’s may be between 50% and 75%.”

Atkins. 2016 Treiber-Kesting. 2013

INTRODUCTION / THEORETICAL BACKGROUND / EXPERIMENT SETUP / SIMULATION / SCENARIO ANALYSIS / PLAN / CONCLUSIONS

(6)

Treiber et al. (2000)

Parameters:

A = maximum acceleration B = comfortable deceleration B_max = maximum deceleration s₀= standstill safe gap

IDM: Intelligent Driver Model

T = speed dependent safe gap = IDM parameter

v₀= desired speed r_i= reaction time

Model characteristics:

 Complete microscopic model

 Transitions between traffic regimes are smooth

 Collision-free

 “Intelligent braking strategy”

 Parameters reasonable interpretation

Cars AVs

A = 3 m/s

²

B = 1.67 m/s

²

B

_max

= 7.5 m/s

²

S

₀

= 2 m

T =1.5 s

= 4

v

₀= norm(30.56;3.5)

r

_i= norm(0.5;1)

Stochastic IIDM Deterministic IIDM A = 3 m/s

²

B = 1.67 m/s

²

B

_max

= 7.5 m/s

²

S

₀

= 2 m

T = 1 s

= 4

v

₀

= 30.56 m/s r

_i

= 0 s

IIDM: Improved Intelligent Driver Model

(7)

of 28

Micro-simulation of a bottleneck in one-lane link with overtake which constitutes a dynamic congestion system

Cars AVs

Stochastic IIDM

Deterministic IIDM

7

Experiment setup

INTRODUCTION / THEORETICAL BACKGROUND / EXPERIMENT SETUP / SIMULATION / SCENARIO ANALYSIS / PLAN / CONCLUSIONS

(8)

Calculate Calculate

Simulatio n

Framewo rk

x_i-1 x_i s_i

v_i v_i-1

L_i

3 hours

DesS LaunchS

Time steps j

�� h

_{� �,}_�

( ^� ^, ^��

� �,�

) ^:

DEMAND INPUT

YE S NO

(9)

of 28

Simulatio n

Framewo rk

9 x_i-1

x_i s_i

v_i v_i-1

L_i

CAR FOLLOWING ALGORITHM

Update equations:

��,�=��−1,�+��,�(∆�)+1 2�

�,�

(∆�)²

�_�,�=�_�−1,�+�_�_,�(∆�)

IID M

(10)

Simulatio n

Framewo rk

x_i-1 x_i

v_i v_i-1

L_i

v_i v_i-1 v_i s_i-1 s_safe s_adv

OVERTAKING PROCEDURE

YE S

�_�,�>�_{�� ,�}+�_{�� }_,�+�_� NO

��≤��_{��}(��(��) ) NO

YE Update S

positions�_�_,�=�_�,�−1+�_{�� }_,�+�_�

�_{�� }_,�

�_�� _�_,_�

��(��)

Calculate Calculate Calculate

(11)

of 28

Simulatio n

Framewo rk

11 x_i-1

x_i s_i

v_i v_i-1

L_i

PERFORMANCE INDICATORS

AVERAGE TRAVEL TIME (ATT)

THROUGHPUT AND CAPACITY: Inflow, Outflow, Capacity drop

QUEUE: upstream vehicles with v<40km/h

DETECTORS: Speed, Flow, Density, Spacing

DRIVING DIRECTIO 13 km N

1 km

40 UPSTREAM

BOUNDARY

DOWNSTREA M

BOUNDARY

ATT

q_in q_out

%C_drop

(12)

Scenario Analysis

Calibration of Base Scenario: 0% AV

Analysis of AVs T and v

₀

parameter sensitivity

Analysis of AVs at different penetration rates Analysis of AVs T and v

₀

parameter sensitivity

Analysis of AVs at different penetration rates

Cars

A = 3 m/s

²

B = 1.67 m/s

²

B

_max

= 7.5 m/s

²

S

₀

= 2 m

T =1.5 s

= 4

v

₀= norm(30.56;3.5)

r

_i= norm(0.5;1)

Stochastic IIDM

Performance Indicators:

 Av. Travel time

 Throughput

 Capacity drop

 Queue

(13)

of 28

INTRODUCTION / THEORETICAL BACKGROUND / EXPERIMENT SETUP / SIMULATION / SCENARIO ANALYSIS 13 / PLAN / CONCLUSIONS

Scenario Analysis: AVs different driving behaviours

Calibration of Base Scenario: 0% AV

Analysis of AVs T and v

₀

parameter sensitivity

Analysis of AVs at different penetration rates

Cars

A = 3 m/s

²

B = 1.67 m/s

²

B

_max

= 7.5 m/s

²

S

₀

= 2 m

T =1.5 s

= 4

v

₀= norm(30.56;3.5)

r

_i= norm(0.5;1)

Stochastic IIDM

Performance Indicators:

 Av. Travel time

 Throughput

 Capacity drop

 Queue

Calibration of Base Scenario: 0% AV

AVs

Deterministic IIDM A = 3 m/s

²

B = 1.67 m/s

²

B

_max

= 7.5 m/s

²

S

₀

= 2 m

T = 1 s

= 4

v₀= 30.56 m/s =110km/h

r

_i

= 0 s

50% Avs

100% AVs

v

₀

80 km/h 110 km/h 145 km/h

T 1.0 s 1.0 s 1.0 s 80 km/h

110 km/h 145 km/h

1.0 s 1.0 s 1.0 s 50% AVs

100% AVs

110 km/h 110 km/h 110 km/h

0.5 s 1.0 s 1.3 s v

₀

110 km/h 110 km/h 110 km/h

T 0.5 s 1.0 s 1.3 s

Analysis of AVs at different penetration rates

(14)

AVs different driving behaviours – AVERAGE TRAVEL TIME v

₀

= Desired speed

parameter sensitivity

T = Speed dependent time gap

parameter sensitivity

1 km

40 UPSTREAM

BOUNDARY DOWNSTREA

M BOUNDARY

ATT

 At 50% AVs: higher v

₀

than the mean of regular cars has no extra benefit.

 At 50% AVs: lower v

₀

than the mean of regular cars could worsen the performance.

 At 50% AVs: high sensitivity of the T parameter on the ATT indicator.

 At 100% AVs: v

₀

key parameter to improve performance.

50% of AVs penetration rate

100% of AVs penetration rate

(15)

of 28

v

₀

110 km/h 110 km/h 110 km/h 110 km/h 110 km/h 110 km/h

T 1.0 s 1.0 s 1.0 s 1.0 s 1.0 s 1.0 s

Scenario Analysis: AVs at different penetration rates

Calibration of Base Scenario: 0% AV

Analysis of AVs T and v

₀

parameter sensitivity

Analysis of AVs at different penetration rates

Cars

A = 3 m/s

²

B = 1.67 m/s

²

B

_max

= 7.5 m/s

²

S

₀

= 2 m

T =1.5 s

= 4

v

₀= norm(30.56;3.5)

r

_i= norm(0.5;1)

Stochastic IIDM

Calibration of Base Scenario: 0% AV

Analysis of AVs T and v

₀

parameter sensitivity

% AVs 0 25 50 75 90 100

AVs

Deterministic IIDM A = 3 m/s

²

B = 1.67 m/s

²

B

_max

= 7.5 m/s

²

S

₀

= 2 m

T = 1 s

= 4

v₀= 30.56 m/s =110km/h

r

_i

= 0 s

Performance Indicators:

 Av. Travel time

 Throughput

 Capacity drop

 Queue

(16)

AVs at different penetration rates – AVERAGE TRAVEL TIME

1 km

40 UPSTREAM

BOUNDARY DOWNSTREA

M BOUNDARY

ATT

14.6

8.1

 Variation of ATT is not linear.

 Significant time savings when

AVs ≥ 50%.

(17)

of 28

AVs at different penetration rates – QUEUE

1 km

40 UPSTREAM

BOUNDARY DOWNSTREA

M BOUNDARY

UPSTREAMBOUNDARY DOWNSTREAMBOUNDARY

 Queue length barely reduced at low market shares.

 Greater change observed from 90%

AVs to 100%AVs.

(18)

Flow – Density

0% AVs 25% AVs 50% AVs 75% AVs 100% AVs 90% AVs

AVs at different penetration rates – FUNDAMENTAL DIAGRAMS

(19)

of 28

source: Sustainable Transport Lab source: Confused.com

AVS IN MOTORWAYS MIXED IN TRAFFIC DEDICATED LANES FOR AVS IN MOTORWAYS

INTRODUCTION / THEORETICAL BACKGROUND / EXPERIMENT SETUP / SIMULATION / SCENARIO ANALYSIS / 19 PLAN / CONCLUSIONS

How / What should we plan?

(20)

Consumer Surplus: economic measure of the (dis)benefits of existing and new users.

 Average Travel Time = f (%AVs)

 VoT:

source: Ministry of Transport of Denmark

 Av. Cost = Av. Travel Time * VoT

 Throughput = f (%AVs)

AVs at different penetration rates – Consumer Surplus

Direct benefits

New benefits

(21)

of 28

 Scaling of results:

‒ 2 peak periods per day

‒ 1 year (285 days)

AVs NORMAL CARS NORMAL CARS NORMAL CARS + AVs

NORMAL CARS+ AVs

NORMAL CARS + AVs

100%

0%

33.3%

�� = 35.2 +0 + 0

�� = 5.5+5.5 +5.5

�� = �� .� �� / ��

�� =�� .� �� / ��

AVS IN MOTORWAYS MIXED IN TRAFFIC DEDICATED LANES FOR AVS IN MOTORWAYS

INTRODUCTION / THEORETICAL BACKGROUND / EXPERIMENT SETUP / SIMULATION / SCENARIO ANALYSIS / 21 PLAN / CONCLUSIONS

AVs at different penetration rates – Consumer Surplus

(22)

Conclusions, limitations and future research

 High uncertainty about AVs driving behavior (at 100% and mixed traffic), difficult to parametrize models.

 At early stages of AVs in mixed traffic, AVs time safe gap (T) is a very sensitive parameter to reduce travel times and increase capacity.

 AVs definitely have the potential to benefit the traffic flow and capacity in congested motorways but it could be difficult to perceive it at early stages. Results show significant benefits when

market penetration rates of AVs are larger than 50%.

 Results suggest it will be more beneficial to plan towards motorways with dedicated lanes for

AVs rather than AVs mixed in traffic. However, several practical problems could arise when

implementing into real-world.

(23)

of 28

Conclusions, limitations and future research

 Limitations of the modelling framework:

‒ Homogeneous motorway segment.

‒ Single lane. Overestimation when expanding results to motorways with more lanes.

‒ No trucks modelled.

‒ Investigation of congestion generated by a permanent bottleneck.

 AVs driving behavior assumptions:

‒ AVs all behave in the same way.

‒ AVs system facilitators behavior. Only overtake if they do not disturb any other vehicle.

 The simulation is developed using the IIDM model for both normal cars and AVs (with different parametrizations). Other microscopic models could be also used and results could vary.

 Work is limited to analyze changes in vehicle dynamics and does not consider potential induced demand.

INTRODUCTION / THEORETICAL BACKGROUND / EXPERIMENT SETUP / SIMULATION / SCENARIO ANALYSIS / 23 PLAN / CONCLUSIONS