1. Introduction
Traffic modeling plays a vital role in the planning, management, and development of road networks [
1]. This requires accurate and realistic characterization of traffic behavior [
2,
3]. Traffic models have been extensively used to mitigate congestion [
4] because it is a key concern in urban areas around the globe. Congestion causes excessive fuel consumption, air pollution, and safety issues, and has an adverse economic impact [
4,
5,
6].
It has been shown that nearly 88 billion dollars were lost due to congestion in the United States in 2019 [
7]. The increasing number of vehicles on the roads adds to congestion. For example, there were 284.5 million cars registered in the United States in 2019, and this increased by 0.84% in 2020 [
8]. Large traffic queues are created during congestion which impede the smooth flow of traffic. Traffic flow is influenced by the time and space required for vehicles to adjust to the environment [
9]. These factors affect driver response and result in velocity differences.
Three types of models are commonly used to characterize traffic: macroscopic, microscopic, and mesoscopic. Macroscopic models consider collective vehicle behavior and are typically used to determine velocity, flow, and density [
10]. Microscopic models consider individual vehicle behavior using parameters such as time and distance headways, position, and velocity [
11]. They are used to predict vehicle dynamics and are often based on driver response [
12]. Mesoscopic models are a hybrid of microscopic and macroscopic models and so share the properties of both [
13].
Pipes [
14] and Reuschel [
15] were the first to introduce microscopic traffic models. Velocity was determined using the distance between following and leading vehicles. However, their models are simplistic and cannot adequately characterize traffic behavior [
16]. Newell [
17] introduced a model which considers distance headway and velocity. With this model, large distance headways can produce high velocities and low-density traffic [
5]. However, it can also create excessive acceleration, which is unrealistic. In [
18], an Optimal Velocity Model (OVM) was proposed which depicts a constant driver response regardless of the traffic conditions. This is not a realistic characterization as traffic and velocity differences are ignored, leading to unstable dynamics [
9]. Moreover, the acceleration can be very high when the velocity is far from the equilibrium distribution, and the density is not considered. It has been shown that this model leads to traffic accidents because of the small distances between vehicles.
Helbing and Tilch [
19] introduced the Generalized Force Model (GFM) which employs negative velocity differences considering following vehicles and the OVM. Unfortunately, it can produce unrealistic results as there are rapid changes in acceleration. This is because only aggressive drivers are considered so slow and typical driver behavior is ignored [
9]. This model was improved in [
20] using both negative and positive velocity differences [
5]. Gipps [
21] developed a different traffic model but the acceleration is not realistic [
16]. In addition, the behavior may not correspond to the model parameters [
22].
Treiber et al. [
22] developed the Intelligent Driver (ID) model which is an improvement of the Gipps model [
21]. Driver behavior is considered along with distance headway and velocity to provide smooth acceleration [
23,
24]. The ID model has been used to avoid collisions during emergencies [
25], and the results are similar to those obtained when observing real traffic [
26]. However, it employs an acceleration constant
which is the same for all traffic conditions, so traffic physics is ignored. The ID model was modified in [
27] for traffic at signalized intersections. However, a value of
was employed, so real traffic conditions were neglected [
28]. This constant was chosen as the best fit for general traffic environments. A similar approach was employed in [
29] for deceleration at intersections. In this case, the distances between vehicles are very small even with a high velocity, which is unrealistic [
9]. The ID model has been incorporated in MovSim to evaluate longitudinal vehicle movement [
25] and also in PTV VISSIM and Simulation of Urban Mobility (SUMO). However, driver behavior is not considered in this model [
30].
One application of the ID model is with Connected and Autonomous Vehicles (CAVs) to improve traffic safety and passenger comfort [
31,
32]. Li et al. [
33] employed this model to characterize the car-following behavior of CAVs, while Schakel et al. [
34] proposed an improved ID model to explore CAV traffic stability. In this case, there is an abrupt decrease in velocity when the density reaches half its maximum (critical density), which is unrealistic as the velocity should be smooth. A modified ID model was proposed in [
35] to better characterize real traffic conditions. However, CAV behavior with this model is not realistic because under actual traffic conditions, the drivers take longer to achieve a smooth car-following behavior and the variations in headway are greater [
36]. The ID model has also been incorporated into cooperative and Adaptive Cruise Control (ACC) systems [
37,
38]. Unfortunately, the safe distance between vehicles with this model is too small when the velocity is high, which can result in accidents when employed in ACC and related systems.
A model is introduced here to realistically characterize traffic flow based on velocity differences and driver sensitivity. This improves the ID model which can produce unrealistic traffic behavior because of the constant acceleration exponent. Driver response to changes in velocity is known as sensitivity. The interval for a vehicle to adapt to variations in traffic conditions is referred to as the distance headway, while the time required to traverse this interval is known as the time headway. Both the proposed and ID models are evaluated for a platoon of 52 vehicles on a circular road of length 2000 m with periodic boundary conditions. The results obtained demonstrate that the proposed model provides better traffic characterization than the ID model.
2. Traffic Models
The ID model is a car-following model based on the behavior of leading vehicles. The acceleration is determined by the velocity, distance headway, and driver response which is
, where
and
are the average and the desired velocities, respectively. The acceleration is [
22]
where
is the maximum acceleration,
is the acceleration constant, and
is the distance headway as illustrated in
Figure 1. The desired distance headway is [
22]
where
is the jam spacing during congestion,
is the time headway,
is the change in velocity, and
is the minimum acceleration.
With the ID model, traffic behavior is determined by the exponent
. However, this fixed value is inadequate for diverse traffic scenarios. Further, it is not based on real vehicle dynamics so poor results can occur. Hence, a variable exponent is proposed which is based on the difference in velocity between forward and rearward vehicles. This difference is given by
where
is the forward vehicle velocity and
is the rearward vehicle velocity as shown in
Figure 2. The product of velocity and density is the traffic flow [
23]
and substituting Equation (4) in Equation (3) gives
A high density results in slow vehicles and a small distance headway
. Conversely, a low density results in a large distance headway between vehicles and large vehicle speeds [
9]. Further, the time headway
is inversely proportional to the flow [
39], so Equation (5) can be expressed as
Driver sensitivity can be expressed as
where
is the safe time headway required to avoid accidents. When
is less than
, vehicles are slow and there is a small distance between them so drivers respond quickly. Conversely, when the time headway is greater than the safe time headway, the flow is smooth so the distance headway is large and driver response is slow. Substituting Equation (7) in Equation (6) gives the proposed acceleration exponent
and replacing
in Equation (1) with Equation (8), the proposed model is obtained as
This model employs the distance and time headways to account for vehicle alignment due to changes in traffic. This is more accurate than the ID model that relies on an arbitrary constant regardless of the traffic conditions. Further, this constant is not based on traffic physics.
Traffic density is the inverse of the distance headway and at steady state
where
is the equilibrium distance headway [
40]. At steady state,
so substituting Equation (2) in Equation (1) and solving for
gives
for the ID model and
for the proposed model. Equation (10) indicates that the ID model headway is based on an arbitrary fixed value
δ and so is unrealistic. On the other hand, Equation (11) is based on actual traffic parameters such as the distance and time headways and so
δ is not a constant.
The steady-state traffic flow is given by
which for the ID model is
and for the proposed model is
This indicates that the steady-state traffic flow for the proposed model incorporates the time and distance headways. With a small headway, driver response is quick as there are significant interactions between vehicles and drivers have little time to react. In this case, the traffic flow is low [
9,
41]. Conversely, with a large headway there are few vehicle interactions and driver response is slow. In addition, drivers have more time to react so the flow is high [
9,
41].
3. Performance Results
The proposed and ID models are evaluated using the Euler scheme [
40] implemented in MATLAB. A circular road of length 2000 m is considered with periodic boundary conditions. The simulation time is 200 s and the time step is 0.5 s. The desired velocity is 30 m/s and the maximum and the minimum deceleration (negative deceleration) are 0.5 m/s
2 and 3 m/s
2, respectively [
29]. The jam spacing is 2 m [
37], and the forward and rearward distance headways are both 25 m [
2]. The time headway varies depending on the traffic conditions and is usually between 0.5 s and 2.6 s [
42]. Here, the forward time headway is 1.5 s, the rearward time headway is 1.6 s, and
is
s. The ID model is evaluated for
s [
6], while the proposed model is evaluated for
and
s. The acceleration exponent varies from 1 to
and is commonly set to 4 [
22], so
and
is considered. There are 52 vehicles on the road, each with a length of 4.5 m [
43]. The simulation parameters are given in
Table 1.
Figure 3 presents the ID model velocity on a
m circular road for
, and
. When
, the initial velocity is
m/s and increases to
m/s at
s. It then decreases to
m/s at
s and then increases to
m/s at
s. When
, the velocity is
m/s at
s, increases to
m/s at
s, and then decreases to
m/s at
s. The highest velocity is
m/s at
s. When
, the velocity increases from
m/s at
s to
m/s from
s to
s. It decreases to
m/s at
s, increases to
m/s at
s, and is
m/s at
s.
The velocity on a
m circular road with the proposed model for
,
,
2, and
s is shown in
Figure 4. When
s, the initial velocity is
m/s and increases to
m/s at
s. It decreases to
m/s at
s and then increases to
m/s at
s. When
s, the initial velocity is
m/s and increases to
m/s at
s. It is
m/s at
s and
m/s at
s. When
s, the velocity increases from
m/s at
s to
m/s at
s. It decreases to
m/s at
s and then increases to
m/s at
s. When
s, the initial velocity is
m/s and increases to
m/s at
s. It decreases to
m/s at
s and then increases to
m/s at
s. When
s, the initial velocity is
m/s and increases to
m/s at
s. It is
m/s at
s and increases to
m/s at
s.
Figure 5 presents the flow with the ID model on a
m circular road for
and
. When
, the initial flow is
veh/s and increases to
veh/s at
s. It is
veh/s at
s and increases to
veh/s at
s. When
, the initial flow is
veh/s and increases to
veh/s at
s. It decreases to
veh/s at
s and then increases to
veh/s at
s and
veh/s at
s. When
, the initial flow is
veh/s, increases to
veh/s at
s, and then decreases to
veh/s at 104 s.
The flow on a
m circular road with the proposed model for
and
s is given in
Figure 6. When
s, the flow is
veh/s at
s, increases to
veh/s at
s, and then decreases to
veh/s at
s. When
s, the flow is
veh/s at
s, increases to
veh/s at
s, and then decreases to
veh/s at
s. When
s, the flow is
veh/s at
s and increases to
veh/s at
s. It decreases to
veh/s at
s and then increases to
veh/s at
s. When
s, the flow is
veh/s at
s and then increases to
veh/s at
s. It decreases to
veh/s at
s and then increases to
veh/s at
s. When
s, the flow is
veh/s at
s and increases to
veh/s at
s. It is
veh/s at
s and increases to
veh/s at
s.
Figure 7 presents the trajectories of a platoon of
vehicles with the ID model on a
m circular road, and the results at
s are given in
Table 2. The thick black line is the trajectory of the
st vehicle while the pink lines show the trajectories of the following
vehicles. When
,
Figure 7a shows that the position of the 1st and
th vehicles is
m and
m, respectively, while the position of the
th and
th vehicles is
m and
m, respectively. When
,
Figure 7b shows that the position of the
st and
th vehicles is
m and
m, respectively, while the position of the
th and
th vehicles is
m and
m, respectively. When
,
Figure 7c shows that the position of the
st and
th vehicles is
m and
m, respectively, while the position of the
th and
th vehicles is
m and
m, respectively.
The trajectories of a platoon of
vehicles with the proposed model on a
m circular road are presented in
Figure 8, and the results at
s are given in
Table 3. The first vehicle is denoted by a thick pink line, while the following
vehicles are shown by black lines. When
s,
Figure 8a shows that the
st vehicle position is
m and the
th vehicle position is
m, whereas the
th and
th vehicle positions are
m and
m, respectively. When
s,
Figure 8b shows that the
st vehicle position is
m and the
th vehicle position is
m, whereas the
th and
th vehicle positions are
m and
m, respectively. When
s,
Figure 8c shows that the
st vehicle position is
m and the
th vehicle position is
m, whereas the
th and
th vehicle positions are
m and
m, respectively. When
s,
Figure 8d shows that the
st vehicle position is
m and the
th vehicle position is
m, whereas the
th and
th vehicle positions are
m and
m, respectively. When
s,
Figure 8e shows that the
st vehicle position is
m and the
th vehicle position is
m, whereas the
th and
th vehicle positions are
m and
m, respectively.
Figure 9 presents the density with the ID model over time and space on a
m circular road and
Table 4 gives the congestion results. When
,
Figure 9a shows that there is congestion from
s to
s as the density is
. At
m, it decreases to
at
s and
at
s. At
s, the density is
at
m and decreases to
at
m. When
,
Figure 9b shows that there is congestion from
s to
s as the density is
. It is between
and 0.50 from
s to
s. At
s, the density is
0 at
m and decreases to
at
m and
at
m. When
,
Figure 9c shows that there is congestion between
.0 m and
m as the density is
and at
m this continues until
s. At
s, the density is
at
and decreases to
at
m and
at
m.
The density with the proposed model over time and space on a
m circular road is given in
Figure 10 and
Table 5 gives the congestion results. When
s,
Figure 10a shows that there is congestion from
s to
s as the density is
. It decreases to
at
m and
s and then increases to
at
m and
s. At
s, the density is
between
m and
m and decreases to
at
m. When
s,
Figure 10b shows that the density is
from
s to
s which indicates congestion. After the congestion dissipates, the density is
at
s and
m, and increases to
at
s and
m. Between
m and
m, the density is
at
s. It increases to
at
m and then decreases to
2 at
m. When
s,
Figure 10c shows there is congestion from
s to
s as the density is
. The density decreases to
at
m and
s and then increases to
at
m and
s. At
s, the density between
m and
m is
and decreases to
at
m. When
s,
Figure 10d shows there is congestion between
s and
s as the density is
. At
m, it decreases to
at
s and
at
s. At
s, the density is
at
m and decreases to
at
m and
at
m. When
s,
Figure 10e shows there is congestion between
s and
s as the density of
. It decreases to
at
m and
s and
at
m and
s. At
s, the density is
at
.0 m and
at
m.
Figure 11 presents the ID model flow for
, and
. When
,
Figure 11a shows that at
.0 m the flow is zero from
s to
s. It is
veh/s at
s and
veh/s at
s. At
m, the flow is
veh/s at
s, and decreases to
veh/s at
m and
veh/s at
m. When
,
Figure 11b shows that at
m the flow is zero from
s to
s. It increases to
veh/s at
s and varies between
veh/s and
veh/s from
s to
s. At
s, the flow is
veh/s at
.0 m and increases to
veh/s at
m where it remains. When
,
Figure 11c shows that at
m, the flow is zero from
s to
s. It increases to
veh/s at
s and varies between
veh/s and
veh/s from
s to
s. At
s, the flow is
veh/s at
m and increases to
veh/s at
m where it remains.
Figure 12 presents the flow with the proposed model on a
m circular road for
, 1,
,
and
s. When
s,
Figure 12a shows that at
m, the flow is zero from
s to
s. It increases to
veh/s at
s and then decreases to
veh/s at
s. At
s, the flow is
veh/s between
m and
m and increases to
veh/s at
.0 m. It then decreases to
veh/s at
m and remains constant. When
s,
Figure 12b shows that at
m, the flow is zero from
s to
s. It increases to
veh/s at
s and is approximately constant until
s. At
s, the flow is
veh/s between
m and
.0 m and increases to
veh/s at
m. It decreases to
veh/s at
m and
veh/s at
m. When
s,
Figure 12c shows that at
.0 m the flow is zero from
s to
s. It increases to
veh/s at
s and is approximately constant until
s. At
s, the flow is
veh/s between
m and
m and increases to
veh/s at
m. It decreases to
veh/s at
m and
veh/s at
m. When
,
Figure 12d shows that at
m, the flow is zero from
s to
s. It increases to
veh/s at
s and is approximately constant until
s. At
s, the flow is
veh/s at
m and decreases to
veh/s at
.0 m and
veh/s at
m. When
s,
Figure 12e shows that at
m, the flow is zero from
s to
s. It increases to
veh/s at
s and
veh/s at
s. At
s, the flow is
veh/s at
m and decreases to
veh/s at
m and
veh/s at
m.
4. Discussion
The results presented in
Figure 3 indicate that the variations in velocity with the ID model increase with the acceleration exponent
. Further,
Figure 4 illustrates that the variations in velocity with the proposed model increase with the time headway.
Figure 5 shows that the variations in flow with the ID model increase over time as
increases. Similarly, the results for the proposed model presented in
Figure 6 indicate that the variations in flow over time increase with a larger time headway.
Figure 7 and
Figure 8 present the vehicle trajectories in time and space with the ID and proposed models, respectively, and the results are summarized in
Table 2 and
Table 3.
Table 2 indicates that as
increases, the distance traveled by the
st and
th vehicles increases, while that traveled by the
th vehicle decreases. However, the distance traveled by the
th vehicle is approximately the same.
Table 3 shows that a decrease in the time headway decreases the distance traveled by the
st vehicle, while the distance traveled by the
th vehicle increases between
s and
s, decreases between
s and
s, and then increases. Between
s and
.0 s, the distance traveled by the
th vehicle increases and then decreases between
s and
s. The distance traveled by the
th vehicle decreases between
s and
s and then increases. These results show that the vehicle positions with the ID model obtained using an arbitrary constant are not based on real traffic conditions. Conversely, the positions according to the proposed model are based on the time headway. This results in a smooth flow with the platoon of vehicles which is more realistic than the ID model.
Figure 9 shows that as
increases, the ID model density increases over time so there is congestion with a large
. Moreover,
Figure 11 indicates that the flow increases with
.
Figure 10 shows that the density with the proposed model is small for a small time headway and vice versa. In addition,
Figure 12 indicates that the changes in flow are proportional to the time headway.
Overall, the results given indicate that the traffic behavior with the proposed model is more realistic than with the ID model. In particular, the flow and velocity are smoother, and the variations in flow and density over time are smaller. The ID model can produce unrealistic traffic behavior such as the significant congestion shown in
Figure 9c. This is because the ID model employs an arbitrary constant whereas in the proposed model this constant is replaced with a variable based on the time headway. The results presented highlight the importance of considering real traffic parameters in traffic flow models to accurately characterize traffic behavior.