Optimal Highway Bottleneck Operation with Active Traffic Management and Connected Automated Vehicle
Advisor: Rahim F. Benekohal
This study focuses on the mitigating impact of traffic incidents on highway operations with variable speed
limits (VSL) and traffic turbulence reduction strategies to alleviate incident-induced congestion. The
emergence of connected autonomous vehicles (CAV) and smart infrastructure, proposed management
strategies (MS) aim to enhance operating speed at the bottleneck, reduce arrival volume, and control
arrival speed during incident response time periods to improve traffic operation.
Responders often need to temporarily close lanes to ensure the safety of all involved parties during
incidents, resulting in a time gap between incident clearance and full traffic restoration. This temporal
difference is measured through recovery attempts at individual sensor locations, with an average of 4.7
attempts contributing approximately 11.6 minutes to low-speed duration for the entire dataset. Recovery
attempts follow specific patterns (U type, nV type, UV type), influencing the number of attempts needed
to restore normal operating speeds. U type recovery averages 4.3 attempts, contributing 13.9 minutes,
while nV type sees 4.4 attempts, adding about 9.1 minutes. UV type recovery requires an average of 6.9
attempts, with each attempt increasing low-speed duration between 5.3 to 18.1 minutes; a low value
corresponds to a high number of attempts, while a high value corresponds to a low number of attempts.
Consequently, low-speed duration can be accurately predicted through rigorous linear regression analysis
and fully connected neural networks (FCNN), with the potential for mitigation through the integration of
ATM and CAV technologies, emphasizes the importance of addressing key influential variables. Variables
such as speed drop, arrival volume, and arrival speed present opportunities for refining incident MS. The
proposed incident MS are VSL, CAVs traveling on open lanes early, and CAVs maintaining their open
lane near the incident site, and they aim to reduce low-speed duration by improving operating speed,
minimizing arrival volume, and controlling arrival speed. The effectiveness of these strategies is assessed
using Vissim simulation software due to challenges in their field implementation.
The effectiveness of the first MS, VSL is contingent upon critical factors such as activation distance,
compliance rates, and posted speed limits. VSL demonstrates the potential to mitigate shockwave
propagation by reducing both arrival speed and arrival rate. This effectiveness is evident in Vissim
simulation results, particularly in terms of travel time, where a consistent reduction is observed as
activation distance increases in all CAV market penetration. An increase in compliance rates of humandriven
vehicles further contributes to a reduction in travel time. The second MS involves CAVs traveling
on open lanes early, aiming to minimize turbulence near the incident site. This strategy is most effective
when implemented from 1 mile upstream of the incident site. The third MS, involving CAVs maintaining
their open lane near the incident site, is not universally beneficial across all CAV market penetrations but
becomes advantageous as CAV market penetration and volume increase.
In tackling the intricate challenges of finding combination of optimal incident MS, this study introduces
innovative data-driven methodologies, decision trees and random forests. These methodologies provide a
transparent and interpretable framework for data-driven decision-making, leveraging crucial variables like
traffic volume, CAV market penetration, VSL activation distance, and VSL compliance rate. The
extensive dataset serves as the foundation for identifying optimal traffic management strategies across
diverse incident scenarios, utilizing a meticulous ranking procedure that considers factors such as
maximum queue length, total travel time, and average throughput during queues. Rankings are capped at
10 for a focused evaluation, with tied ranks resolved based on priorities like maximum queue length, total
travel time, and average throughput during queues.
Results from both decision trees and random forests show promise, revealing a consistent trend of
improvement across all CAV market penetration, including low levels like 10%. These data-driven
strategies contribute to a reduction in queue length (19.24% and 19.63%), travel time (17.86% and
17.76%), and an increase in average throughput during queues (5.74% and 5.70%). Even at low CAV
market penetration 10%, the MS obtained from data-driven strategies showed significant improvements in
queue length, travel time, and throughput, which is a promising for future incident management.
