Article citation information:
Nguyen, X.H., Vu, T.V.A., Than, Q.V., Pham, V.T., Nguyen, T.A., Ha, M. Applying
neural network techniques to determine traffic flow redirection proportions in
road networks. Scientific Journal of Silesian
University of Technology. Series Transport. 2025, 127, 267-275. ISSN: 0209-3324. DOI: https://doi.org/10.20858/sjsutst.2025.127.16
Xuan-Hien NGUYEN[1], Thi Van Anh VU[2], Quoc
Viet THAN[3],
Viet Thanh PHAM[4],
The Anh NGUYEN[5],
Muon HA[6]
APPLYING NEURAL NETWORK TECHNIQUES TO DETERMINE
TRAFFIC FLOW REDIRECTION PROPORTIONS IN ROAD NETWORKS
Summary. This article explores traffic
management strategies for addressing unpredictable events in transportation
networks, focusing on situations where road segment capacity is reduced due to
factors like traffic accidents or disruptions. The research aims to determine
the proportion of traffic flow redistribution needed to maintain network
efficiency under such conditions. A novel method is proposed to mitigate
congestion by rerouting vehicles from heavily loaded roads, identified by high
network load coefficients, to alternative routes. The approach also calculates
the optimal volume of redirected traffic to avoid overloading other parts of
the network, thereby minimizing the risk of secondary congestion. To achieve
this, neural network-based survey and regression analysis techniques are
utilized, offering precise and data-driven solutions for traffic redirection.
The study highlights the potential of improving urban traffic flow through
enhancements to indirect traffic control systems integrated into Intelligent
Transportation Systems. By optimizing vehicle rerouting strategies, the
proposed method seeks to increase ITS efficiency, especially in scenarios with
high congestion risks or traffic accidents. This approach promises a more
resilient and adaptive urban transportation network, ensuring smoother traffic
operations and reduced congestion impacts.
Keywords: traffic management, intelligent transportation
systems, intelligent neutral network, traffic flows.
1. INTRODUCTION
In
the practice of traffic management, a number of different measures are used to
solve the problems of low efficiency of the road transport network. The
solutions proposed and tested in real conditions achieve certain results
depending on the object under study and are able to improve various indicators
of the road transport network. Among them, Intelligent Transportation System
(ITS) is a widely used means for traffic management in large cities [1, 2].
Overcoming the limitations of this system in solving traffic congestion
problems can simultaneously lead to an improvement in the quality of traffic
functioning in many densely populated cities around the world.
Studies
on traffic management highlight that redirecting traffic flow is accomplished
through the efficient operation of the ITS and its indirect traffic flow
management subsystem [3, 4, 5]. Research findings indicate that the success of
the indirect traffic flow management approach largely depends on how well the
proposed method aligns with the behavior and preferences of road network users
(drivers).
Studies
[8, 9] determine the volume of vehicle traffic within a transportation network
by assessing the capacity of a given road section relative to its traffic
intensity. The key metric for this evaluation is the remaining capacity of the
route. Based on this value, traffic management decisions establish the maximum
number of vehicles that can be redirected to the route.
The
proportion of traffic flow redirection in the network depends not only on the
capacity of individual road sections but also on drivers' route choices [6].
Therefore, traffic management scenarios must be designed to align with drivers’
preferences and decision-making processes. This alignment is achieved by
calculating and implementing road measures that reflect the priorities and
behaviors of road users when selecting routes during traffic management.
To
gather insights into driver route preferences, a survey method was employed.
The collected data was then analyzed using a neural network approach, allowing
for a deeper understanding of drivers' behavior and enabling the development of
more effective traffic management strategies.
2. RESEARCH METHOD
The route
choice of traffic participants during a traffic accident is influenced by
numerous factors. As noted in [7], a driver's decision is shaped by both
subjective and objective aspects of the transportation system. These include
the distance to the destination, travel time, familiarity with the proposed
route, organizational features of movement along that route, direction of
travel, and final destination. However, the impact of these factors on the
driver’s decision varies and is determined by the significance coefficient
assigned to each factor.
It is
important to highlight that this significance coefficient is not fixed; it
fluctuates depending on the driver’s physical and psychological condition, as
well as external influences at the time of decision-making. This variability
underscores the complexity of modeling and predicting
driver behavior under such circumstances.
In this
study, Vissim is used to evaluate the
throughput of an organized intersection with traffic lights [10]. Vissim modeling is used as
a tool to determine the relationship between the values of traffic light
cycles, the duration of the traffic light permissive phase, the intensity and
speed of the traffic flow and the number of traffic lanes of the roadway (Fig.
1).
The
conventional model of the street and road network consists of one main multi-lane
road, five adjoining roads and one bypass road, which is parallel to the main
road, having a capacity equivalent to the main road (among the roads considered
adjoining the main road). Vehicles on the main road move along 4 lanes. Each
adjoining road has two traffic lanes directed to the right relative to the main
direction of traffic on the road. The redistribution of traffic flow from the
main roads to the bypass road is carried out at intersections through a system
of dynamic information boards and road signs, which better facilitates the
process of redistribution and informing road users.
Fig. 1. The diagram of
the studied reference model of the transport network in Vissim
A study was
conducted to investigate the factors influencing drivers' decision-making in
situations with a high risk of congestion. A survey of 111 drivers of various
groups, including women and men, professional and private drivers, and
representatives of different age and experience categories, provided valuable
data on drivers' preferences and priorities when choosing a route [Fig. 2]. The
main conclusion drawn from the analysis of the survey results is that drivers'
choices depend on the relative differences between different routes, rather
than on the characteristics of each route individually. That is, drivers
evaluate the "benefit" of the differences between the proposed
routes.
The study
examined four distinct traffic conditions (A1, A2, A3, A4), represented
by changes in the network load coefficient (z). These scenarios included
the following cases: z < 0.5, 0.5 < z < 0.7, 0.7 < z
< 0.9, and z > 0.9. To enhance comprehension, the variations in
the z coefficient were visually depicted through changes in the color scheme, reflecting the assessment of traffic
conditions. This visualization allowed participants to evaluate road scenarios
more effectively.
a) b)
c)
d)
Fig. 2. Questionnaire
for choosing a scenario for moving through networks of vehicle drivers with
different values
a) z < 0.5;
b) 0.5 < z < 0.7; c) 0.7 < z < 0.9; d) z >
0.9
3. RESULTS AND DISCUSSION
The results presented in the Tab. 1
and Fig. 3 indicate that as traffic conditions worsen, the dominant (selected)
route becomes increasingly attractive compared to other routes. This preference
is primarily due to a reduction in travel time, despite a slight increase in
mileage. In these scenarios, the dominant route consistently represents the one
where the reduction in the cost of movement for road users approaches its
maximum value as z→1.
Tab. 1
The result of the survey on the
choice of scenario for
movement through the networks of vehicle drivers
|
Choice of drivers (%) |
Route No1 |
Route No2 |
Route No3 |
Route No4 |
|
z < 0.5 |
72 |
18 |
14 |
8 |
|
0.5 < z < 0.7 |
14 |
26 |
13 |
12 |
|
0.7 < z < 0.9 |
2 |
12 |
6 |
6 |
|
z > 0.9 |
12 |
44 |
67 |
74 |
Fig. 3. The result of the survey on
the choice of scenario for
movement through the networks of vehicle drivers
As the z value increases, drivers initially consider alternative
routes. However, a transition to the dominant route occurs as it offers the
most favorable balance of reduced travel cost, even
under heightened traffic conditions. This dynamic highlights the interplay
between travel time, mileage, and cost-efficiency in route selection under
varying traffic loads.
The survey results and their analysis reveal that drivers’
decision-making regarding traffic patterns is influenced by multiple factors
acting simultaneously. To uncover general trends in drivers' route choices, the
regression method is essential for determining the impact of each factor on
decision-making behavior. The use of neural network
methods is justified by prior studies highlighting their effectiveness in analyzing complex, multifactorial data.
Artificial neural networks are employed to process the survey results and
identify patterns in drivers’ route choices. The structure and parameters of
the neural network are tailored to the specific characteristics of the survey
data. For this analysis, Matlab-Simulink
2021b software is utilized, enabling effective modeling
and interpretation of the data (refer to Fig. 4 and 5). This approach ensures a
robust analysis of the interplay between various factors influencing route
selection.
Fig. 4. The structure of the
constructed neural network
Fig. 5. Results of the survey data
processing process
The following parameters define the
neural network constructed based on the analysis of the survey results.
The overall regression coefficient
reached R = 0.879, with a maximum value of R = 0.986 observed in
the test sample. These findings demonstrate that the constructed neural network
effectively models the desired relationships.
The training results show that the
correspondence coefficient between the input parameters and the forecasted
outcomes exceeds 87%, which is considered an acceptable level. The observed
differences between the neural network’s predictions and the survey results can
be attributed to the complexity of the input parameters, variations in driver behavior when choosing routes, and differences in traffic
network structures, even under identical traffic conditions.
This can be attributed to the social
characteristics of the respondents, which do not conform to a universal pattern
but are significantly influenced by the individual subjective factors unique to
each respondent.
To verify the achieved results, a
neural network is used to determine the proportion of vehicles that should be
redirected in situations where the coefficient z > 0.9 at a traffic
intensity of 4500 vehicles/hour. 4 traffic management scenarios with the values
of traffic flow and trip characteristics shown in Tab. 2. The results (the
proportion of each scenario chosen by road users, %) are also presented in the
table.
Tab. 2
Result of applying a neural network to
determine the share of
traffic flow redirection through the network
|
Traffic flow diagram |
Load level |
Reduction in travel
time (min.) |
Extension of overrun (km.) |
Improving the quality of service |
Shares of
choice(%) |
|
No1 |
z > 0.9 |
26 |
1.41 |
3 |
80.71 |
|
No2 |
z > 0.9 |
22 |
1.15 |
1 |
7.58 |
|
No3 |
z > 0.9 |
17 |
1.17 |
1 |
1.54 |
|
No4 |
z > 0.9 |
15 |
1.18 |
1 |
1.09 |
In the presented congestion
situation (z > 0.9), the proportion of traffic sce-narios
chosen by the participants and offered to drivers is determined using the
neural network algorithm (as shown in the table) and is consistent with the
observed patterns of choice in the previous survey. The total number of choices
that do not reach 100% - reflecting the inconsistent randomness of drivers'
choices, which are influenced by factors not considered in this survey
(familiarity with the traffic route, personal predictions of traffic
conditions, drivers' personal routes,...) – accounts for 9.08%.
The direction of such a volume of
vehicles in modern conditions is carried out by components of the intelligent
transport system, including indirect traffic management subsystems, which can
be implemented using dynamic information boards or different types of road
signs.
4. CONCLUSION
When
traffic accidents or other unforeseen incidents take place, leading to the
temporary closure of one or more traffic lanes and a substantial reduction in
the capacity of a road section, congestion inevitably develops within the
transportation network. This congestion disrupts the normal flow of vehicles,
increases travel time, and can have broader economic and environmental
consequences. To minimize these adverse effects, appropriate traffic management
strategies must be implemented. These strategies involve redirecting vehicles
onto alternative routes that have sufficient capacity to handle the additional
traffic, thereby pre-venting further congestion from forming on these roads.
Advanced
traffic control systems, particularly those utilizing neural network methods,
can play a crucial role in optimizing the redistribution of traffic flow. By analyzing real-time traffic data, driver behavior, and habitual route choices under congested
conditions, these systems can determine the most efficient way to divert
vehicles. This ensures a smoother traffic flow across the transport network,
reducing delays, improving overall road efficiency, and enhancing commuter
experience.
References
1.
Zhankaziev S.V. 2010.
Concept of creating an intelligent transport system on federal highways.
NIR. 83 p.
2.
Nguyen X.H. 2018. „Status of road traffic, ITS
existing solutions and level of development of intelligent transport systems in
Vietnam”. Avtomobil'. Doroga.
Infrastruktura 2(16). 9 p.
3.
Vu T.V.A. 2024. „Prevention of traffic congestion with
the help of subsystems of directive and indirect traffic flow management”. Bulletin
of MADI 4(79): 82-88.
4.
Buslaev A.P. 2003.
Probabilistic and simulation approaches to the optimization of road traffic.
Russian Academy of Sciences V. M. Prikhodko. Moscow: Mir. 368 p.
5.
Gazvan A.H. 2006.
„International models for assessing the level of road safety”. Science and
technology in the road industry 3: 3-9.
6.
Efimenko D.B. 2012. „Methodological foundations of
building navigation systems of dispatch control of transportation processes on
road transport (e.g. urban passenger transport)”. Diss. doctor of technical
sciences. 05.22.08. Мoscow. 479 p.
7.
Kosolapov A.V. 1992. „Increase of efficiency of
information support of road traffic participants in the cities”. Diss.
candidate of technical sciences. 05.22.10. Мoscow.
178 p.
8.
Kocherga V.G. 2001. „Estimation and forecasting of the
road traffic parameters in the intellectual transportation systems”. Rostov
n/D: Rost. State Construction University. 130 p.
9.
Utkin A.V. 2006. „Modeling
of driver behavior and quality assessment of a mixed
traffic flow”. "Organization and traffic safety in large cities"”. In:
Collection of reports of the 7th International Conference. St.-Petersburg.
P. 84-86.
10. Xuan Can Vuong,
Rui-Fang Mou, Trong Thuat Vu, Hoang Van Nguyen. 2021.
„An adaptive method for an isolated intersection under mixed traffic
conditions in Hanoi based on ANFIS using VISSIM-MATLAB”. IEEE Access 9:
166328-166338. DOI: 10.1109/ACCESS.2021.3135418.
Received 05.03.2025;
accepted in revised form 09.05.2025
Scientific Journal of Silesian University of
Technology. Series Transport is licensed under a Creative Commons
Attribution 4.0 International License
[1] Faculty of Automobile Technology, Hanoi University of
Industry (HaUI), 298 Cau Dien street, Hanoi, Vietnam. Email: hien.nguyen15@haui.edu.vn. ORCID:
https://orcid.org/0000-0002-1552-1867
[2] Faculty of traffic police, People’s s police university, 36 Nguyen
Huu Tho street, 7th district, Hochiminh city,
Vietnam. Email: anhvu7587@gmail.com.
ORCID: https://orcid.org/0009-0000-4948-1616
[3] Faculty of Automobile Technology, Hanoi University of Industry
(HaUI), 298 Cau Dien street, Hanoi, Vietnam. Email: viettqcnoto@haui.edu.vn. ORCID: https://orcid.org/0000-0002-2320-4386
[4]
Faculty of Automobile Technology, Hanoi University of Industry (HaUI), 298 Cau
Dien street, Hanoi, Vietnam. Email: thanhpv@haui.edu.vn.
ORCID: https://orcid.org/0000-0002-2472-2871
[5]
Faculty of Automobile Technology, Hanoi University of Industry (HaUI), 298 Cau
Dien street, Hanoi, Vietnam. Email: anhnt_cnot@haui.edu.vn. ORCID:
https://orcid.org/0009-0009-7250-0382
[6]
Faculty of Information Technology, Telecommunications University, Nha Trang,
Vietnam. Email: muon.ha@mail.ru. ORCID:
https://orcid.org/0000-0003-4385-1916