Abstract

The paper under the name “ A dynamic traffic assignment model for highly congested urban
networks” written by Moshe E. Ben-Akiva , Song Gao, Zheng Wei , Yang Wen proposes a
study of enhancing and calibrating DynaMIT-P, a mesoscopic traffic simulation system for a
18-square-mile area in Beijing, which included 1698 nodes and 3180 directed links. The
model simulated approximately 630,000 vehicles over 4 hours during the morning peak, using
data from 2927 origin-destination pairs. This paper presents those following sections: the first
is the introduction, the second part is about the modelling challenges and solutions, the third is
a case study and finally the fifth gave conclusion and future directions.

The complexities and solutions in modelling highly congested urban traffic networks are
explored in the paper with a focus on Beijing, China. It highlights the necessity for advanced
Dynamic Traffic Assignment (DTA) models that can realistically simulate dynamic traffic
flows and manage demand in intricate urban environments. The paper identifies several
challenges, including handling extensive networks with numerous directional links, closely
spaced on- and off-ramps, severe congestion, and significant interference from non-motorized
traffic. Initial calibration of a DTA model for Beijing faced issues with unrealistic queue
formations and inaccurate sensor data. Solutions presented include the use of the Path-size
Logit (PSL) model for improved route choice modelling and a refined lane-group queuing
model, which collectively enhance simulation accuracy and calibration effectiveness. The
paper underscores the advantages of simulation-based DTA models over analytical models,
noting their ability to capture detailed traffic dynamics and evaluate operational strategies
such as personalized traffic information and dynamic tolling. It also discusses recent
advancements in DTA models, including better data structures, algorithms, and distributed
computing, which improve both realism and computational efficiency. Ongoing challenges
include accurately replicating congestion, modelling complex intersections and short links,
and dealing with interference from non-motorized traffic, emphasizing the critical need for
precise calibration with historical data.

Dynamic Traffic Assignment (DTA) models often use route choice models to translate origin-
destination flows into path and link flows. The study highlights the limitations of the
Multinomial Logit (MNL) model, which assumes independent and identically distributed
error terms, leading to unrealistic congestion in urban networks with overlapping paths.
Specifically, in the Beijing network, MNL models over-predicted expressway usage, resulting
in excessive congestion. To address this, the study explores alternatives like the C-Logit
model and the Path-size Logit (PSL) model. PSL introduces a correction term based on path
size to account for overlapping paths, improving the accuracy of route choice predictions.
Implementing PSL significantly reduced gridlock and enhanced traffic flow realism in
simulations of the Beijing network.

Roads in DTA models are often represented as links connected by nodes (intersections). Links
can be divided into segments to reflect capacity variations due to geometric changes like lane
adjustments. Each segment typically includes a high-speed "moving part" and a "queuing
part" where stop-and-go traffic occurs. Accurate queuing models are vital for realistic
congestion estimation. While segment-based queuing models are simpler and efficient, they
can misrepresent traffic dynamics by allowing queues to form indiscriminately across lanes,
affecting traffic flow accuracy, particularly in complex intersections. To address these issues,
lane group-based models were developed, grouping lanes by their specific functions (e.g., left
turn, through, right turn) to better account for lane restrictions and traffic behaviour. This
approach reduces unrealistic congestion and bottlenecks by aligning queuing dynamics with
actual lane usage, as demonstrated in the Beijing network study.

2 3 Short links
Short links in complex networks, like Beijing’s, pose challenges for traffic simulations. Early
studies showed excessive congestion due to unrealistically slow vehicle movement and
unnecessary queuing on these links. These issues highlight the limitations of mesoscopic
supply models, particularly on short links. The discussion addresses these problems
separately, focusing on segment and lane group inaccuracies.
he study addresses issues in traffic simulation models related to short road segments and lane
group capacity. Short segments in networks, like those in Beijing, often cause unrealistic
congestion due to the speed-density relationship used in mesoscopic models. Specifically,
vehicles on these short segments may decelerate excessively due to the rapid increase in
density. To resolve this, the calibration of minimum speeds for short segments was adjusted to
reflect average observed speeds. Additionally, the study tackles problems with lane group
acceptance capacities, which can lead to over-predicted congestion, especially in congested
networks. The acceptance capacity was improved by using a modified formula that ignores
acceptance capacity constraints when there is no queue. The study also introduces dynamic
output capacities to account for time-dependent impacts of non-motorized traffic on vehicular
flow, improving the model's accuracy in urban environments.

3. Case study
This case study applies the DynaMIT-P system, a simulation-based Dynamic Traffic
Assignment (DTA) tool, to model traffic in a highly congested urban network in Beijing.
DynaMIT-P integrates a microscopic demand simulator and a mesoscopic supply simulator,
enabling efficient simulation of traveler behavior and vehicle movements under various traffic
conditions. The study focuses on Beijing's West 2nd Ring Road and its surrounding areas,
notorious for severe traffic congestion. The network model comprises 1698 nodes and 3180
links within an 18-square-mile area, simulating approximately 630,000 vehicles during the
morning peak hours. Data from Remote Traffic Microwave Sensors (RTMS) and GPS-
equipped taxis were used to calibrate the model, which aimed to capture complex traffic
dynamics and network performance under extreme congestion. The study faced challenges
due to Beijing's unique traffic patterns, including unusually short road links and the potential
for gridlock.
The calibration of the DynaMIT-P model for Beijing's highly congested urban network
required adjusting 69,093 variables, including 46,832 time-dependent OD flows and speed-
density parameters across 3180 segments. Calibration was conducted using the Simultaneous
Perturbation Stochastic Approximation (SPSA) algorithm, formulated as a constrained
minimization problem. The model was calibrated against real-world data, including traffic
counts and travel times, with Root Mean Square Error (RMSE) and Normalized Root Mean
Square Error (RMSN) used to evaluate accuracy. The calibration process successfully
reflected the network's high congestion levels, with improvements in the objective function
and stable RMSN values, indicating the model's suitability for simulating traffic conditions in
Beijing.
The DynaMIT-P Dynamic Traffic Assignment (DTA) system, once calibrated, can effectively
evaluate traffic management strategies. This study demonstrates its application by assessing
the short-term benefits of Beijing's "Rotating No-Driving Day" restriction, implemented post-
2008 Olympics to reduce pollution and traffic congestion. This policy restricts private car
usage one weekday per week based on license plate numbers, significantly reducing demand.
The analysis compares simulated traffic scenarios with and without the restriction. Key
findings include a 20% decrease in demand leading to an 8.4% reduction in vehicle arrivals at
destinations and a 17.9% decrease in average travel time. Additionally, queuing times on links
with significant congestion were halved. The results indicate that such restrictions can
substantially enhance road efficiency and mitigate congestion in the short term.

Conclusion and future direction

This study enhances traffic simulation in congested urban areas by integrating key features
into DynaMIT-P: an advanced route choice model, explicit lane group representations, short
link handling, and dynamic capacities for non-motorized traffic. Applied to Beijing, the model
was calibrated with traffic sensor and floating car data, demonstrating its capability to
evaluate management strategies effectively. Future improvements will focus on refining the
route choice model with vehicle trajectory data and enhancing accuracy with better network
coding and surveillance data.