www.nature.

com/scientificreports

OPEN Research on highway traffic

flow prediction model
and decision‑making method
Yuyu Zhu 1, QingE Wu 2* & Na Xiao 1

In order to solve the problem of traffic congestion in a certain area, this paper develops a set of traffic
optimization decision system. For analyzing the actual traffic conditions and calculating the traffic
volume, density and traffic speed, a traffic prediction model is established and updated iteratively to
modify the prediction model parameters. Based on this model, the congestion degree is estimated
at the current road section, thus, an intelligent decision-making and the coordinated optimization
methods are proposed. Moreover, this paper implements some application experiments on the
isometric road of a three-intersection and obtains better prediction results of traffic density and traffic
speed on the three-section highway. At the same time, compared with other existing prediction
methods, the prediction model presented in this paper not only has higher accuracy, shorter
prediction time and stronger anti-interference ability, but also has better effect on vehicle diversion.
In addition, it also greatly relieves the traffic pressure on the road, maximizes the complementary
advantages between intersections, and balances the good cooperation between each intersection.

With the rapid development of economy and society, urban traffic congestion has become increasingly seri-
ous, For coordinating networked vehicles and traffic lights at urban intersections, Rodriguez et al.1 proposed a
decentralized control framework. However, the framework didn’t discuss the different gradient to dynamically
coordinate the vehicle’s planned intersection arrival time. For better highway event detection, traffic analysis
and traffic forecast,2 proposed a method that could reduce delays by 55–70 percent in different traffic conditions.
However, the highway traffic flow model was actually a high-order nonlinear time-varying system, which makes
the state estimation of the model a very difficult problem. In order to achieve accurate estimation of traffic flow,
a traffic prediction optimization model system is proposed in this paper.
In fact, in order to solve the problem of traffic congestion and congestion, the main method of some ­studies3–5
was to establish mathematical models for traffic flow to control and optimize the entire traffic system. For solv-
ing some problems of urban traffic control and management, Jiahui et al.6 proposed a dynamic and deadline-
oriented road pricing mechanism model to open up a new solution to the management and control of traffic.
Although these studies had advanced methods, the effect of control optimization was not satisfactory due to
the randomness and complexity of the traffic system, especially in some megacities where road congestion and
traffic accidents occur frequently. For optimized design of some technical parameters, Junwei et al.7–9 established
the traffic flow cellular automata model based on the construction strength, anti-collision ability and mechani-
cal properties of expressway traffic safety facilities. However, these studies did not solve a real complex various
system and the limitations of the traffic flow model.
For a good optimization on construction parameters, Qiu et al.10 proposed a construction quality control
method for expressway traffic safety facilities based on parameter fusion tracking control and piecewise linear
regression analysis of construction schedule, but the mode cannot meet the needs of urban traffic. In addition,
Kumar et al.11 proposed an effective traffic control system based on ant colony algorithm and Internet of vehi-
cles technology, but this management method is not suitable for all industries. To avoid traffic congestion, Sun
­Danying12 proposed an intellectual administrative system of traffic events. However, it turned out that the system
could not entirely ensure the smooth road conditions and safe driving. Kuppusamy et al.13 proposed a novel
intelligent traffic control framework, but it was not effective for complex congestion. For large and complex traffic
conditions, Wang et al.14 put forwards a more accurate trunk line control system based on the concepts of red
wave and green wave, however, the traffic pressure and relieve congestion reached only 70% at the intersection.
For most advanced traffic simulation and evaluation, Louati et al.15 developed Pannal, which could effectively

1
Faculty of Engineering, Huanghe University of Science and Technology, Zhengzhou 450000, China. 2School of
Electrical and Information Engineering, Zhengzhou University of Light Industry, Zhengzhou 450002, China. *email:
[email protected]

Scientific Reports | (2022) 12:19919 | https://doi.org/10.1038/s41598-022-24469-y 1

Vol.:(0123456789)
 www.nature.com/scientificreports/

Calculation of Establishment of
Iterative update of Intelligent decision
parameters for traffic state prediction
model
state model

Experiment verification,
result analysis and
comparison

Figure 1.  Flow chart of traffic flow prediction and decision method.

solve the problem of traffic congestion and facilitate the emergency vehicles. At the same time, Yang et al.16 pro-
posed an improved near-end policy optimization algorithm, but they ignored the relationship between vehicles
and traffic lights, so the effect of urban traffic control is not comprehensive enough. For traffic signal control,
Huang Mingxia et al.17 proposed an intelligent traffic system based on FPGA traffic signal light, and Tajdari et al.18
proposed a new adaptive control method based on the reference of unknown time-varying basic graph model.
However, the previous studies did not give a solving method for high order nonlinear model of traffic congestion.
To solve the high order nonlinear model of traffic congestion, this paper proposes the model linearization
iterative updating method and develops a traffic prediction and decision system. The system consists of four
parts: (1) An analysis of actual traffic conditions, that is, calculation of traffic volume, density and speed; (2)
Establishment of traffic flow state prediction model, that is, congestion degree, influence of future time period on
surrounding road branches, and state estimation modeling; (3) Iterative update of model, that is, in the applica-
tion of traffic flow analysis, studying the traffic flow data under various conditions and modifying the parameters
of the prediction model until the requirements are met; (4) Intelligent decision-making, that is, game algorithm
in regional roads, traffic optimization, regional scheduling, diversion of traffic congestion and smooth roads.
Figure 1 shows the framework of proposed methods and system construction. These studies provide theoretical
ideas and decision-making support for government construction, urban planning and road length and width
design, as well as target detection and tracking methods for places and departments using video stream.
Article highlights:
Simultaneously, the novelty of present work had been abstracted as follows:

• A traffic coordination optimization method is proposed.

• A traffic flow prediction model is established.
• An intelligent decision-making method is given, and a coordinated optimization method for regional traffic
is also given.
• A set of traffic optimization decision system is developed.

Calculation of parameters on traffic flow condition

According to works of Payne and Papa Georgiou et al.1,2, the model of traffic flow is discussed as follows:
Take an interval segment with intersection in a region for example. Divide the segment appropriately into L
segments, each of which i is spatial sampling length. Traffic detectors are set at the beginning and end of the
whole road to provide measured data as the input of the dynamic model, including traffic flow qi (k) and average
speed vi (k). The detection period T is 60 s, usually a cycle of 10–60 s. The density of traffic flow ρi (k) is calculated
from the traffic flow qi (k) and average speed vi (k), and these parameters based ­on1,2 are calculated as follows:
qi (k) = αρi (k)vi (k) + (1 − α)ρi+1 (k)vi+1 (k) (1)

T 
(2)

ρi (k + 1) = ρi (k) + qi−1 (k) − qi (k) + ri (k) − si (k)
�i

T
vi (k + 1) = vi (k) + {v[ρi (k)] − vi (k)}
τ
Tξ γ T ρi+1 (k) − ρi (k) (3)
+ vi (k)[vi−1 (k) − vi (k)] −
�i τ �i ρi (k) + 
where
      
b
v(ρ) = vf exp − 1 b ρ ρcr (4)

In addition, in the formula, ri (k) is the traffic flow rate in the i th paragraph at the entrance at time kT , and si (k)
is the traffic flow rate at the exit. The parameter vf represents the free speed of traffic flow, namely, the maximum
speed of current traffic flow. ρcr is the density of traffic flow when the traffic flow reaches the maximum, namely,

Scientific Reports | (2022) 12:19919 | https://doi.org/10.1038/s41598-022-24469-y 2

Vol:.(1234567890)
www.nature.com/scientificreports/

the critical density, and b, τ , γ , ξ , , α is the adjustment coefficient of the equation. Equation (4) is a steady-state
traffic flow model, which describes the velocity density relationship of steady-state traffic flow.

Establishment of prediction model

According to the analysis of high-speed traffic flow data in various situations, these measurement data can be
established. The dynamic model is carried out multiple tests by simulation and experiments based on the above
calculation of the parameters of the traffic flow dynamic model. Moreover, according to the principle of auto-
matic control nonlinear system, the state of traffic flow and its measurement values can be obtained at the k time:
x(k + 1) = f [x(k)] + Ŵ[x(k)]w(k) (5)

y(k + 1) = h[x(k + 1)] + v(k + 1) (6)

where w(k) and v(k) are the zero-mean noise vectors, and
E[w(k)] = 0, E[w(k)w ′ (j)] = Q(k)δkj , E[v(k)] = 0, E[v(k)v ′ (j)] = S(k)δkj .

Q(k) is a variance matrix of the system noise, S(k) a measure noise matrix, Ŵ[x(k)] is the noise drive matrix,
which these can be obtained by experiments and measures. f [x(k)] and h[x(k)] are two mapping function of state
and measure mapping function to perform the state estimation and measure value, respectively. According to
the calculation of positive definite matrix, f [x(k)] and h[x(k)] are given by repeated experiments and reasoning
as follows:
M
− 12 (y(k)−x(k))T Q−1 (k)(y(k)−x(k)) (7)

f [x(k)] = ce i=1

M
 S2 (k)
h[x(k)] =  T   (8)
i=1 S2 (k) + y(k) − x(k) Q−1 (k) y(k) − x(k)

where c is an adjustment constant, and M is the measurement number of sensors.

To define a following norm on distance, if there is a certain state value x̂(k) meet to the following formula for
the corresponding measurement value y(k), x̂(k) is the established state of traffic flow at the k time.
 
x̂(k + 1) − y(k + 1) < δ (9)
where δ is a given threshold.

Iterative update of prediction model

In order to make the developed system predict traffic flow state more accurately, a new traffic prediction method,
called iteration, is presented here. The iterative update algorithm (IUA) needs to integrate information of differ-
ent state models, and update state, state error, measurement error, process gain value K(k + 1) and covariance
accordingly P(k + 1).

Iterative update of model. Further, according to the above formula (5)–(9), the state value x̂(k) can be
updated by a gain learning rate K(k + 1) as follows:

(10)
 
x̂(k + 1) = x̂(k) + K(k + 1) y(k + 1) − x̂(k + 1)

where K(k + 1) is an adjustable gain rate. It can be obtained through the incremental change of the state function
and the measurement function.
According to the calculation of covariance, the state covariance can be updated finally to zero, i.e.,
 
∂h
P(k + 1) = I − K(k + 1) P(k) (11)
∂ x̂(k + 1)
∂f
where I is a unit matrix, ∂ x̂(k) and ∂ x̂(k)
∂h
are a partial derivative matrix, respectively.
When the initial filter value x̂(0) and the initial filtering covariance matrix P(0) are acquired, the recursive
algorithm is calculated in the following procedure.
x̂(0), P(0) → P(1) and x̂(1) → K(1) and x̂(1) → x̂(2) and P(2) → ……, which can be calculated step by step to
estimate the system state. Obviously, the recursive equations all have to be implemented online.
In order to make the prediction model more accurate in application, the state of the above traffic flow model is
constantly updated by using the interaction of states of each section. It is necessary to calculate the corresponding
partial derivative matrix in the recursive equation and obtain the expression consistent with the above recursive
Eq. (11), so that online recursive calculation can be carried out to obtain the estimated value of system state at
each moment.
(i)
If the probability update of state transitions is assumed as µk , then the final prediction of state interactive
update is:

Scientific Reports | (2022) 12:19919 | https://doi.org/10.1038/s41598-022-24469-y 3

Vol.:(0123456789)
 www.nature.com/scientificreports/

Estimates vs. True state

40

35

30

25

veh/m
20

15

10

True state
5
New algorithm IUA estimate
EKF estimate
0
10 20 30 40 50 60 70 80
t/m

Figure 2.  Comparison of estimation of IUA and EKF algorithms for traffic flow density.

�  �(i) (i)
x = xk|k µk (12)
k|k
i

From the definition of variance and mean, the total covariance of corresponding state error is updated as:
 (i) � �(i) � �(i) (i)
Pk|k = [Pk|k + E( x − xk|k )( x − xk|k )′ ]µk (13)
k|k k|k
i

Application of model. In here, an interval with an intersection is discussed. For each road in the intersec-
tion, only three equal length sections are discussed: one entrance, one driving section, and one exit.
Only the entrance section of the first road section and the exit section of the third road section provide detec-
tion data of vehicle flow and average speed as the input and output measurement values of the state estimation
model. Take the flow at the head of the first section q0, the average velocity v0 as the input value. Take the density
at the end of the third section ρ3, the average velocity v3 as the output. The purpose of state estimation is to
accurately estimate the value of traffic variable in the section without detector according to the measured value
of detector through constant state updating operation. During the experiment, it is assumed that the initial state
value is the given normal value.
According to the model equation based o ­ n1,2, the traffic volume of each section is expressed by the density
and speed ρi (k) and vi (k) at three-way intersection, where i = 1, 2, 3.

Comparison between iterative update and existing algorithm.

1. Initial conditions

Simulate the above models. Assume the traffic density is uneven that is caused by sudden traffic accidents on
the first road segment. The simulation parameters in the model are based on the values i­ n2, i.e., vf = 60 (km /h),
ρcr = 20 (veh/m), b = 3, τ = 19.4/3600, γ = 34.7, ξ = 1,  = 1, α = 0.95,  = 0.8 km.
′
Give initial state vector is: x0 = 10 10 10 0 0 0 ,

 
10 0 0 0 0 0
 0 10 0 0 0 0 
 0 0 10 0 0 0 
 
Give covariance matrix is: P0 =  .
 0 0 0 0.1 0 0 
 0 0 0 0 0.1 0 
0 0 0 0 0 0.1
′
Model probability of state transition is: µ0 = 0.6 0.2 0.2 .


2. Comparison of simulation results of the two algorithms

Using the iterative update algorithm (IUA) proposed in this paper and the existing EKF a­ lgorithm19, compare
the results of the estimation of traffic density and traffic speed on the three-section viaduct highway within 90 s

Scientific Reports | (2022) 12:19919 | https://doi.org/10.1038/s41598-022-24469-y 4

Vol:.(1234567890)
www.nature.com/scientificreports/

Estimates vs. True state

140
True x
EKF estimate
120
IUA estimate

100

v(km/h)
80

60

40

20

0
0 10 20 30 40 50 60
t/m

Figure 3.  Comparison of estimation of IUA and EKF algorithms for traffic flow speed.

Comparison of IUA and EKF on prediction probability

1

0.9

0.8

0.7
probability

0.6

0.5

0.4

0.3
IUA
EKF
0.2
0 10 20 30 40 50 60
times

Figure 4.  Comparison of IUA and EKF algorithms on prediction probability of traffic flow.

Prediction algorithms Prediction accuracy (%) Prediction speed (ms) Anti-interference ability
IUA algorithm proposed 92.36 2.08 Strong
EKF algorithm 78.82 4.19 Weak

Table 1.  Comparison of proposed IUA and existing EKF algorithms.

of simulation, as shown in Figs. 2 and 3. Figures 2 and 3 show the estimation of density and speed on traffic flow
by IUA and EKF algorithms, respectively, which a predicted value is relative to the true value. From two figures,
the estimation of IUA algorithm is better than that of EKF algorithm. IUA estimation curve basically coincides
with the actual truth curve, while the EKF algorithm is relatively poor.
In addition, to evaluate the accuracy of the proposed IUA algorithm, it can be defined as follows:
Assume the total number of experiment statistics is N , the number n of correct prediction is counted by the
counting program, then the probability prediction P of the precision is defined as
n
P= (14)
N

Scientific Reports | (2022) 12:19919 | https://doi.org/10.1038/s41598-022-24469-y 5

Vol.:(0123456789)
 www.nature.com/scientificreports/

For 200 statistical experiments, repeat 60 times, and use the counting program to count the effective responses
n at each times. Then, the probability precision is calculated by formulas (14), as shown in Fig. 4.
Compare the accuracy of the IUA algorithm with the existing better EKF algorithm, the experimental results
show that the accuracy of the IUA algorithm is 92.36%, and that of EKF algorithm is 78.82%, as shown in Fig. 4
and Table 1. In addition, the prediction speed of IUA algorithm is 2.08 ms. However, that of EKF is 4.19 ms.
From Fig. 4 and Table 1, the correct prediction probability of IUA algorithm proposed in this paper is higher
than that of the existing EKF algorithm, and the running time is smaller. By comparison, it is found that the inter-
active algorithm proposed in this paper has more accurate prediction estimate, faster prediction speed, stronger
anti-interference, better effect of congestion evacuation, and solves the problem of road congestion. The method
proposed in this paper not only greatly alleviates the traffic pressure, but also maximizes the complementary
advantages between intersections, thereby balancing the cooperation between each intersection.

Intelligent decision
Based on the video of the current moment and the decision-making system, drivers are informed of the pos-
sible highway condition, a smooth route near the driving expressways, and even the smooth exit and entrance
to a clear highway. In the design of decision system of highway traffic, the traffic state of each section of the
highway must be used as feedback information to determine the control quantity such as speed limit, entrance
regulation rate and information release. Here is a new approach to intelligent decision making called inferential
decision-making.

Decision rules. Let the traffic condition set be X = {x1 , x2 , · · · xn }, decision sets Y = y1 , y2 , · · · ym , so a
 
mapping R from X to Y is an optimal decision rule. For example, if xi is clear of traffic, then decision yj is dij to
proceed; or if xi is traffic congestion, then decision yj is to change the degree of driving on the current expressway
road to dij , and so on. Where dij = µX (xi ) ∧ µY yj ,µX (xi ) is the probability that xi is a member of a given set
 

of conditions X ,µY yj is the probability that yj is a member of the given set of states Y . Then, this decision rule
 

is described as follows:
If a condition x is A, then decision y is B",
If x is A′ , then decision y is B′.
It can be defined as B′ = A′ ◦ (A → B). That is, conclusion B′ can be obtained by combining A′ with the infer-
ence relation from A to B.Here, A is the condition set of the state, and B is the decision set based on condition
A. The above rules can be simply described as follows:
It is known that when A and B, the output is C , that is, there are inference rules.
IF A AND B, THEN C.
To find what the decision output C ′ should be when A′ and B′.

Establishment of decision model.

1. Basic symbol and definition

To establish the regional traffic decision model, the following symbols and definitions are introduced:

(i) Qi (t) denotes the vector of the number of vehicles waiting at intersection i at time t .
Qi (t) = Qi,E (t), Qi,S (t), Qi,W (t), Qi,N (t) . Where Qi,E (t), Qi,S (t), Qi,W (t), Qi,N (t) represent the num-
 
ber of vehicles in the east, south, west and north directions waiting at intersection i at time t , respectively.
(ii) Qi represents the vector of the threshold of vehicle number at the intersection i ,
Qi = Qi,E , Qi,S , Qi,W , Qi,N , where Qi,E , Qi,S , Qi,W , Qi,N indicate the thresholds of the number of

vehicles waiting in different directions, respectively, which can be modified according to the actual situ-
ation.
(iii) S represents the set of all possible strategies or actions in the decision. All feasible strategies of a decision
are a finite set, where all strategies are rule sets. Assume S = {s1 , s2 , · · · , sn }, each si is the following rule
shown in “decision model”.

2. Decision model

s1:  If Qi,E (t) > Qi,E and Qi,W (t) > Qi,W , but Qi,S (t) < Qi,S and Qi,N (t) < Qi,N , then go straight in east and
west direction, after 10 s, make a double left turn until 30 s to satisfy the red light of the east and west turns
direction.
s2:  If Qi,S (t) > Qi,S and Qi,N (t) > Qi,N , but Qi,E (t) < Qi,E and Qi,W (t) < Qi,W , then go straight in north and
south direction, after 10 s, make a double-left turn from north to south until 30 s to satisfy the red light in
north–south left turn direction.
s3:  If Qi,E (t) < Qi,E and Qi,W (t) < Qi,W and Qi,N (t) < Qi,N , but Qi,S (t) > Qi,S , then it is necessary to turn left
to the north and northeast for traffic.
s4: If
 Qi,E (t) > Qi,E and Qi,W (t) > Qi,W , but Qi,S (t) > Q i,S and Qi,N (t) > Qi,N , at the same time,
Qi,E (t) − Qi,E  > Qi,S (t) − Qi,S  an d Qi,E (t) − Qi,E  > Qi,N (t) − Qi,N  an d Qi,W (t) − Qi,W  >
  
Qi,S (t) − Qi,S  and Qi,W (t) − Qi,W  > Qi,N (t) − Qi,N , then go straight in east and west direction, or
     
north–south vehicles can choose a nearby intersection to divert to other unblocked expressway roads.

Scientific Reports | (2022) 12:19919 | https://doi.org/10.1038/s41598-022-24469-y 6

Vol:.(1234567890)
www.nature.com/scientificreports/

Figure 5.  Game tree.

s5: If
 Qi,E (t) > Qi,Eand Q i,W (t) > Qi,W and
 Qi,N (t) > Qi,N , but  (t) < Qi,S, simultaneously,
 
Qi,E (t) − Qi,E  <
 Qi,S
Qi,N (t) − Qi,N  and Qi,W (t) − Qi,W  < Qi,N (t) − Qi,N  and Qi,S (t) − Qi,S  < Qi,N (t) − Qi,N , then turn
  

left to the south and southwest vehicles pass, or both east–west and north-bound vehicles can choose the
nearby intersection to divert to other unblocked expressway roads.

…

The above rules are common: since there are two cases of traffic flow and threshold in each direction, and
the four directions are considered at the same time, there is 24 total of six cases similar to s1, s2 and s3. Then the
difference between the traffic flow in each direction and the threshold should be considered in the remaining
24 − 6 = 10 cases. However, there are 4 differences. When each difference is compared with 3 differences in the
other direction, 3 cases appear 3 × 10 = 30 differences are obtained. Then, there are 30 + 6 = 36 cases altogether.
Therefore 36 rules are defined and 36 elements are contained in S , that are si i = 1, 2, . . . , 36.

Decision coordination. At a certain intersection at a certain time, if a certain rule si is better than any other
rule to make the expressway road smoother that is, the dredging time of the intersection is shorter, then rule si
wins and is denoted as si∗. The coordination model is the game algorithm as follows.

Coordination model. A game with n rules is described as G = {S, U}, where S is the rule set and U is the advan-
tage obtained by comparing two rules. For example, the advantage is that the expressway roads are smoother,
or the intersection is cleared in shorter time, or congestion is cleared, and so on. If m advantages exist, set
U = {u1 , u2 , . . . , um }. If strategy si∗ in this problem is a Nash equilibrium, then it must satisfy
uj si∗ , sic ≥ uj (si , sic ) (15)
 
∀si ∈ S

where, si∗ represents the strategy chosen of the i th rule; sic represents the strategy of all rules except i ; uj is the j th
advantage. Or reorder by advantage, it can get a balance under Nash equilibrium.

Coordination algorithm. The whole traffic coordination process in the region is divided into three levels. The
lower level is the coordination between the intersection and its adjacent intersection. The middle layer is the
coordination between regional sections and intersections. The upper layer is the coordination between the area
segment and its adjacent area segment.

Step 1 If the number of vehicles queuing at intersection 1 exceeds the threshold, a request is sent to adjacent
intersection 2. In here, the threshold is obtained through a large number of congestion experiments, and
then set by the algorithm program in the established model.
Step 2 Adjacent intersections respond to the request and build a game tree as shown in Fig. 5. The letters on
the branches of the game tree represent the rule strategy, and the expressions in the block diagram represent
the advantages of the rule comparison. Only three groups of advantage values are given in the figure, and
other analogies are used to search the game tree and find the balance point according to Eq. (15).

Scientific Reports | (2022) 12:19919 | https://doi.org/10.1038/s41598-022-24469-y 7

Vol.:(0123456789)
 www.nature.com/scientificreports/

Figure 6.  Developed hardware device.

  The highways studied in here are not only urban roads and expressways, but also expressways outside the
city, including mountain expressways. The intersections are in not only urban roads intersections, but also
three intersections of expressways on-ramps and off-ramps, such as Fig. 5.
Step 3 If equilibrium exists, the action strategy of the decision is the strategy when equilibrium is achieved.
Each decision controls the road mouthpiece according to this strategy, and the coordination ends. If there
is no equilibrium, a request is made to the decision at that intersection.
Step 4 Regional decision making responds to the request, carries out game coordination for the intersection
decision under its jurisdiction, and seeks the equilibrium point. If the equilibrium point does not exist, the
regional decision making sends a request to the neighboring decision making.
Step 5 The decisions of adjacent regions respond to the request, carry out game coordination, and seek the
equilibrium. If the equilibrium point does not exist, the coordination fails, then each decision keeps the
original strategy unchanged.

For this developed decision model, a hardware device is configured to provide navigation services for guid-
ance vehicles on traffic congestion section, as shown in Fig. 6. Moreover, the software running environment is
Matlab platform for proposed algorithms. Each communication interface protocol is written by Python. The
hardware is constructed by using microelectronic components to develop an embedded system under Qemu
equipment, as shown in Fig. 7.

Comparison between proposed reasoning decision and existing methods. Calculation of com-
parison index. To evaluate the decision effect, the correct decision probability Pc is to treat two discriminant
results as the sole probability. Assume Nl = max {nil }, where nil is the times of all examined sections participat-
i
ing in matching test rules at time l , respectively. Nc is the number of correct decisions. So there are
Nc
Pc = (16)
Nl
Obviously, the above result is different for different l , but as l increases,Pc will approach a certain stable value.

Simulation results and analysis. The simulation was carried out in two cases. In case 1, the medium density
traffic environment was simulated, and the target of entering the public area was 50 batches. Scenario 2 simulates
the dense traffic environment, and the target of entering the public area is 100 batches. During the simulation,
the random number seed was selected as 15, and the collection period was 4 s. Figures 8 and 9 show the correct
decision rates at two cases 1 and 2 after 60 simulations, respectively.

Performance analysis. As can be seen from Figs. 8, 9, the correct decision rate in dense environment is slightly
lower than that in medium density environment. In addition, the decision of reasoning in the overall perfor-
mance is better than the existing Agent method, because it not only has good decision-making effect, fast pro-
cessing speed, but also strong anti-noise ability. More than 90% accuracy of decision is achieved in the environ-
ment with dense target. In the medium density environment, the correct decision rate of the inference decision
method is 91.28%, 8% higher than that of the Agent method. The processing speed is 0.0936 s, and the Agent
method is 0.0780 s, both less than 1 s.
Compared with Agent, the biggest advantage of the reasoning decision method in this paper is that it can be
applied not only to the dense target environment, but also to the larger system with model transformation and

Scientific Reports | (2022) 12:19919 | https://doi.org/10.1038/s41598-022-24469-y 8

Vol:.(1234567890)
www.nature.com/scientificreports/

Video data acquisition

Video Key Feature extraction

program

network
card Windows (64Bit)
Host

User program

network
card

Generate data External communication

Parameter State Intelligent Coordinated
calculation fusion decision-making optimization

PCIE

Qemu6.2.0 Virtual
Can machine (64Bit)
RapidIO Linux operating system
equipment

Figure 7.  Developed traffic prediction and decision control model.

Correct decision rate (case 1)

1

0.8

0.6
Pc

0.4

0.2 Reasoning
Agent
0
0 5 10 15
t
Figure 8.  Correct decision rate in case 1.

Scientific Reports | (2022) 12:19919 | https://doi.org/10.1038/s41598-022-24469-y 9

Vol.:(0123456789)
 www.nature.com/scientificreports/

Correct decision rate (2)

1

0.8

0.6

Pc
0.4

0.2 Reasoning
Agent
0
0 5 10 15
t
Figure 9.  Correct decision rate in case 2.

Mean of correct decision Computing speed Ability to Adapt to Parameter Setting

Algorithms rate P c (Case1) (Case1) P c (Case2) anti- noise environment Requirements
Inference Decision 0. 9128 0.0936 s 0.9012 Strong Dense target Yes
Agent 0.8265 0.0780 s 0.7755 Weak Sparse target No

Table 2.  Comprehensive comparison between inference decision method and Agent method.

transmission error. But too many parameters of system and the complicated setting are its disadvantage. While
Agent is good at universality and coordination, it is too sensitive and unsatisfactory in anti-noise ability. Another
disadvantage is that once a node is changed, many other nodes will be affected.

Comprehensive comparison. To evaluate the comprehensive performance of each algorithm, a comprehensive

comparison is conducted based on the correct decision rate, calculation speed, anti-noise, adaptability to envi-
ronment and parameter setting requirements. With the quantitative and qualitative methods, the comparison
evaluates the advantages and disadvantages of reasoning decision method and agent method. Table 2 shows the
comprehensive comparison results.
The correct average decision rate in Table 2 refers to the average of 15 time steps of each algorithm in the given
simulation experiment environment 1 and environment 2 after 60 simulation experiments. They are, in fact, the
spatially and temporally average of the rate of correct decisions, and therefore the overall average of the rate of
correct decisions. The calculation speed is the calculation time used by the algorithm to calculate 15 time steps
in the first simulation environment. It is only the calculation time of the algorithm itself. The computer used for
simulation is Intel I7, 4-core, main frequency ≥ 2.2 GHz, 128 GB memory, and the programming language used
is Matlab. The anti-noise capability in Table 2 is determined by adding some disturbance to the model parameters
and some disturbance estimation to the sampling function of the algorithm. The adaptive environment in the
table refers to the environment where the algorithm is applicable, and the parameter setting requirement refers
to that the algorithm contains some parameters set according to the actual environment.

Conclusion
This paper provides dynamic model state estimation from actual highway traffic data thus making it feasible to
monitor traffic flow, which is extended to longer highway sections by dividing them into sections and calculating
each dynamic parameter separately. These tests had been realized in practical traffic application on three-way
intersection.
When the model parameters exceed 10 dimensions and there are more than two kinds of noise interference,
the accuracy of the proposed model is not very good, and further optimization of the model design will be
required in the future. At the same time, according to the actual control needs and the setting of sampling time,
the number of state quantities will be determined by the maximum time required for each calculation of state
under certain circumstances. Thus, the maximum number of expressways between two detectors on the highway
can be roughly estimated, which will actually offer guidance on saving expenses in the process of transportation
equipment installation, and cut down on consumption of resources.

Data availability
All data generated or analyzed during this study are included in this published article. We do wish to share our
raw data, and these data are original.

Scientific Reports | (2022) 12:19919 | https://doi.org/10.1038/s41598-022-24469-y 10

Vol:.(1234567890)
www.nature.com/scientificreports/

Received: 21 July 2022; Accepted: 15 November 2022

References
1. Rodriguez, M. & Fathy, H. Vehicle and traffic light control through gradient-based coordination and control barrier function safety
regulation. J. Dyn. Syst. Meas. Contr. 144(01), 1104–1115 (2022).
2. Papageorgiou, M. Applications of Automatic Control Concepts to Traffic Flow Modeling and Control 35–41 (Spring, 1983).
3. Youssef, N. M., Mehdi, N., Mohamed, B. & Mohammed, E. Multivariate deep learning approach for electric vehicle speed forecast-
ing. Big Data Min. Anal. 4(1), 56–64 (2021).
4. Fan, W. et al. Privacy-aware traffic flow prediction based on multi-party sensor data with zero trust in smart city. ACM Trans.
Internet Technol. 2022, 1–19. https://​doi.​org/​10.​1145/​35119​04 (2022).
5. Azidine, G., Younes, A., Mourade, A. & Ahmed, A. Mathematical validation of proposed machine learning classifier for hetero-
geneous traffic and anomaly detection. Big Data Min. Anal. 4(1), 18–24 (2021).
6. Jiahui, J., Xiaoxuan, Z., Biwei, W., Jinghui, Z. & Yuxiang, W. A dynamic and deadline-oriented road pricing mechanism for urban
traffic management. Tsinghua Sci. Technol. 27(1), 91–102 (2022).
7. Junwei, Z. et al. Freeway traffic flow cellular automata model based on mean velocity feedback. Phys. A 562(15), 1–11 (2021).
8. Junwei, Z., Yongsheng, Q., Fan, Y., Leipeng, Z. & Dejie, X. A multi-value cellular automata model for multi-lane traffic flow under
lagrange coordinate. Comput. Math. Organ. Theory 28(2), 178–192 (2022).
9. Junwei, Z. et al. Expressway traffic flow under the combined bottleneck of accident and on-ramp in framework of Kerner’s three-
phase traffic theory. Phys. A 574(4), 1–11 (2021).
10. Qiu, Y. J., Xiong, L. & Misao, A. Construction technology and quality control measures for expressway traffic safety facilities. Adv.
Transp. Stud. 2(12), 3–14 (2018).
11. Kumar, P. M. et al. Ant colony optimization algorithm with internet of vehicles for intelligent traffic control system. Comput. Netw.
144, 154–162 (2018).
12. Sun, D. Design and implementation of traffic incident warning and release device under Internet of Vehicles. Dig. Technol. Appl.
39(12), 201–203 (2021).
13. Kuppusamy, P., Kalpana, R. & Rao, P. V. V. Optimized traffic control and data processing using IoT. Clust. Comput. 22(01),
2169–2178 (2019).
14. Wang, Z. et al. Coordinated traffic control of urban trunk lines to alleviate congestion of large events. Traffic Inf. Saf. 38(06), 38–45
(2020).
15. Louati, A., Louati, H. & Nusir, M. Multi-agent deep neural networks coupled with LQF-MWM algorithm for traffic control and
emergency vehicles guidance. J. Ambient. Intell. Humaniz. Comput. 11(11), 5611–5627 (2020).
16. Yang, J., Zhang, J. & Wang, H. Urban traffic control in software defined internet of things via a multi-agent deep reinforcement
learning approach. IEEE Trans. Intell. Transp. Syst. 22(06), 3742–3754 (2020).
17. Huang, M. et al. Design of intelligent traffic signal light based on FPGA. Modern Electron. Technol. 44(16), 179–182 (2021).
18. Tajdari, F. & Roncoli, C. Adaptive traffic control at motorway bottlenecks with time-varying fundamental diagram. IFAC-Pap.
OnLine 54(02), 271–277 (2021).
19. Luo, Z., Wu, Z. & Han, Z. Application of extended Kalman filter to the freeway traffic flow model. Acta Autom. Sin. 28(1), 90–96
(2002).

Acknowledgements
This work is supported by Key Science and Technology Program of Henan Province (222102210084); Key Sci-
ence and Technology Project of Henan Province University (23A413007); Center Plain Science and Technol-
ogy Innovation Talents (194200510016); Science and Technology Innovation Team Project of Henan Province
University (19IRTSTHN013), respectively.

Author contributions
Q.W. and Y.Z. wrote the main manuscript text. The photograph in Fig. 5 was taken by Q.W., as well as prepared
Figs. 1, 2, 3 and 4 and 8 and 9. N.X. prepared figures 6 and 7. All authors reviewed the manuscript.

Competing interests
The authors declare no competing interests.

Additional information
Correspondence and requests for materials should be addressed to Q.W.
Reprints and permissions information is available at www.nature.com/reprints.
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and
institutional affiliations.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International
License, which permits use, sharing, adaptation, distribution and reproduction in any medium or
format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the
Creative Commons licence, and indicate if changes were made. The images or other third party material in this
article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the
material. If material is not included in the article’s Creative Commons licence and your intended use is not
permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from
the copyright holder. To view a copy of this licence, visit http://​creat​iveco​mmons.​org/​licen​ses/​by/4.​0/.

© The Author(s) 2022

Scientific Reports | (2022) 12:19919 | https://doi.org/10.1038/s41598-022-24469-y 11

Vol.:(0123456789)