Routing strategy using local information based on a two layer cellular automaton model

: A reliable and efficient traffic network model is required to study urban traffic congestion, which
has increasingly become a global concern recently. The model can then be used to emulate different policy
scenarios to assess different mitigation strategies, which will be very useful to policy- and decision makers.
In this work, we develop a physical traffic model that can be used to investigate the intrinsic property of city
traffic under different human decisions and driving behaviors. Cellular automaton is one of the most
commonly used traffic network model. At its simplest stage, however, it can only model a one-dimensional
problem. The Biham-Middleton-Levine (BML) extends the capability of the cellular automaton model to
model a two-dimensional traffic network problem. However, it can only model two directions: rightward and
downward. Directly modeling the four directions of traffic using the BML model causes jamming and
gridlock problem. Thus, the existing traffic network models have not been sufficiently capable of modeling
the traffic situation realistically. In this work, we propose a two-layer network modeling to address this
intrinsic gridlock problem, where each layer is modeled based on the cellular automaton approach. The model
is developed on a two-dimensional L
2
-square lattice system, and users can specify the vehicle density prior to
running the simulation. A moving strategy is then derived for each vehicle based on the origin and destination
cell locations, where the shortest path is typically assumed. One of the key purposes of this work is to present
the city traffic in a physical way to investigate the inertial characteristics of city traffic and to help bridge the
gap between the simplified cellular automaton models and the complexity of real-world traffic. As such, we
incorporate driving behavior modeling into the two-layer network system by introducing a flexibility index.
Essentially, this index denotes the probability that a driver can deviate from the predefined shortest path when
congestion occurs, i.e., when the next cell in its intended moving direction is occupied by another vehicle.
The rationale behind this feature is that in real situation, drivers have the option to take an alternative path.
We perform a number of traffic simulations to demonstrate the derived model and to gain insight into the
effect of flexibility on the overall traffic flow. In particular, we vary the lattice size (by varying L) and the
traffic density ρ, which will determine the number of vehicles to be simulated. Each vehicle is assigned a
random origin-destination pair, and the corresponding moving strategy is then determined. By plotting the
average vehicle speed as a function of vehicle density, we can find the phase transition point, where the
traffic changes from a free-flow state to a congestion state. Our results show that when drivers are more
flexible, the onset of congestion state is delayed to a higher density value. In other words, for the same
density value, introducing a higher flexibility results in a higher average speed. This suggests that the
vehicles can reach the destination faster, even if they need to cover a longer travel distance. This hypothesis is
confirmed as we observe the effects of flexibility on the total distance traveled and the total number of
completed journeys. Overall, our simulation results are consistent with the real traffic situations. This model
can be further extended to mimic the traffic network more realistically by introducing more complexity in the
system (e.g., the system lattice layout). As such, we will be able to evaluate some scenarios that the existing
traffic models can not emulate accurately, such as the effect of traffic disruptions on the overall network flow.