Analysis and Simulation of Traffic Flow

Temo Bardzimashvili

December 01, 2003

Abstract
In this project we considered the heterogeneities (namely the abrupt changes in speed) occurring in traffic flow and their consequences. As the simulation base we introduced follow-the-leader model. Then after the simulation results were obtained we compare them to the theoretical results based on the nonlinear wave equations which were used to describe traffic jams.

Work done in partial fulfillment of the requirements of Michigan State University MTH 843, advised by Professor MacCluer.

Table of Contents
Introduction Theoretical Background Simulation Results and Discussion Summary and Future Work References Appendix A Appendix B 1 1 2 3 5 6 7 8

1. Introduction
Today, when cars are part of every day life and more and more families tend to own more than one car, the traffic congestion becomes a very important issue. Traffic problems on highways and in urban areas attract considerable attention. Many scientists have tried to develop various mathematical models (such as the car following model, the cellular automaton model, the particle-hopping model) in order to describe and, subsequently, optimize traffic flow. These models describe diverse situations with different assumptions and simplifications. This project is constructed as follows. We considered the simplest approach to the traffic flow theory the microscopic car following model, which describes the two-car case. We then used this model to perform a simulation using simulation toolbox in MATLAB. We compared obtained results to the different macroscopic traffic models. In this project we considered one-lane, one-direction traffic.

2. Theoretical Background
We started with very typical and simple model of traffic flow so called follow-the-leader model. The idea of this approach is that the vehicle is affected only by the vehicle ahead of it. The follow-the-leader model proposes the following rule for the traffic flow [1]:
x n (t ) ( x n 1 (t ) x n (t ))

(1)

where t is time, is the reaction time and xn is the position of the nth car. The so-called sensitivity coefficient may depend on different parameters including the distance between cars and the speed if nth car. Referring to [2] we can say that for ? we have the stable traffic flow, while for >? the traffic flow is unstable. In (1) model a driver tries to adjust speed (decelerate or accelerate) depending on the difference in speeds (or if the distance between cars is too small). It must be mentioned that equation (1) agrees with the optimal velocity model proposed by Newell in 1961 [3]:

x n (t
where V is the optimal speed.

) V ( x n 1 (t ) x n (t )) ,

-1-

3. Simulation
We simulated the follow-the-leader model described in section 2 using the simulation software Simulink. Since car speed is affected by the behavior of a car ahead, we can infer that the first car determines behavior of the others. The whole system can be represented by the following control diagram (Figure 1):

Figure 1. The scheme of follow-the-leader model system.

Here V is the desired speed, is the sensitivity coefficient, the acceleration and V is the actual velocity difference.

is the reaction time, a represents

The transfer function of this system shown on Figure 1 is expressed by the following equation:

U V

e s

At first we considered the simplest system consisting of only two cars. The Simulink model diagram is shown on Figure 5 in Appendix A. Using this model we experimentally checked stability/instability conditions for the follow-the-leader models. We checked different values for and (we concentrated our attention on < ? and >? combinations of these two parameters). The simulation did show that the instability condition can be expressed as > ? . Then we assumed that the traffic consists of n number of equally spaced cars. Appendix B shows the full simulation scheme of the follow-the-leader model for n cars (n > 2). On Figure 6 the main diagram of the system is shown. Also, Figure 7 represents the subsystem used for calculation of headways and relative velocities. This system receives the bus with k absolute velocities (or absolute distances) and after subtraction of velocities (distances) it sends k-1 relative velocities or headways to the outlet bus. We initiated our system by setting up initial velocities and initial absolute positions of 20 cars. We started with the equal car speeds and equal headways. Then, to add some disturbance to the system, we used the pulse generator and subtracted signal produced by it from the constant speed of the leading car. This can be explained as a short, unexpected drop in the speed of the leading car and its immediate recovery. We expected the following cars to respond to this event and adjust their speeds. We tuned up our system by choosing such values for and , so that our system would be stable. Then we calculated and recorded the time propagation of headway distance (actually, though we used only the headway distance we also recorded relative and absolute velocities and the absolute distance).

-2-

Also we changed the values of and , in order to obtain the unstable system. Then we repeated the simulation.

4. Results and Discussion

The simulation was performed for stable system ( =0.01 and =0.8). The simulation results for the stable system are shown on Figure 2 and Figure 3.

Figure 2. The spatio-temporal evolution of the traffic density.

Figure 2 shows the propagation of the density of the in both time and space. The density in this case is defined as the total number of cars divided by the headway: N k xk where N=20 for our system. Figure 3 shows the snapshots of the density distribution for five different times (300, 500, 1000, 1500 and 1800 seconds respectively). The solid and dashed lines were introduced for the better distinction of the consecutive snapshots.

-3-

Figure 3. The density distribution in space for the five different times.

The simple approximation we made in section 2 gives us good illustration of the processes running in the system. One can see that the density wave occurring after the impulse travels through the traffic (from the first car to the last one) and fades away. At the same time the rate of fading decreases as the wave propagates. To describe the traffic jams as the density waves the non-linear wave equations were proposed. Among the most typical equations the Burgers equation was used. The solution of this equation yields the picture of the density wave behavior described on Figure 4 [3].

Figure 4. Relaxation process of the nonuniform flow to the uniform steady flow in the stable traffic region above the coexisting curve [3].

This figure shows that the density wave occurs in the nonuniform traffic flow and propagates from the head to the tail of the traffic gradually fading. The initial density profile has the socalled kink antikink wave form which is described by the Burgers wave equation. -4-

Besides the Burgers equation, the Korteweg-Devries (KdV) and Modified Korteweg-Devries (MKdV) nonlinear equations can be used to describe traffic jams. They are usually used to describe another type of nonlinear waves called solitons. They occur due to the abrupt impulse (i.e. the change of homogeneity of the system) and their main property is that they can travel long distances with minor changes in the wave height [3]. In the conclusion to the simulation part it must be mentioned that the stability threshold ? observed for 2 car follow-the-leader model was shifted for 20 car follow-the-leader model. For the last the stability threshold was about 1.7.

5. Summary and Future Work

In this project we considered such basic mechanisms of traffic flow as follow-the-leader and fluid models. We performed the simulation of the traffic consisting of equally spaced cars with the same starting velocities. Modeling traffic flow is a complex problem and it always leaves an extensive space for the further research. In order to create more realistic model we can consider more advanced version of follow-the-leader model where we can introduce the concept of the desired distance. In this model drivers would adjust their car speeds according to both differences in velocities and between the relative and desired distances. The new model can look like:
a (t ) l[ Dr (t ) Dd (t )] * v(t ) ,

where besides the variables described above, l is the gain for the relative distance and Dt(t) and Dd(t) are the relative and desired distances respectively. Furthermore, inhomogeneous car distribution can be taken into account (i.e. the distances between cars are not equal). Also, the deeper analysis of the KdV and MKdV equations can be performed in order to describe traffic jams.

-5-

6. References
[1] The MATLAB Notebook v.1.5.2, A Model of Traffic Flow [Link] [2] C. R. MacCluer, Industrial Mathematics , Prentice Hall, 2000 [3] Takashi Nagatani, The physics of traffic jams [Link]

-6-

Appendix A

Figure 5. The simulation diagram for 2 car system.

-7-

Appendix B

Figure 6. The main simulation diagram for n car system (n > 2).

-8-

Figure 7. The subsystem used in Speed Differences and Distance Differences subsystems of the main simulation scheme for n car system (n > 2).

-9-