Performance Improvements for Large-Scale Traffic Simulation in MATSim - Computational Approaches for Urban Environments - page 208

After this brief description of the MATSim model, in the following section the
different traffic micro-simulation models available for MATSim are presented in
more detail.

9.2.2 Traffic Simulation Model

Traffic flow simulations range from detailed physical simulations to macroscopic
models. Detailed physical simulation attempt to capture as many traffic flow
phenomena as possible, e.g., car following and lane changing, by representing space
continuously and simulating very small time steps, e.g., Fellendorf and Vortisch
( 2010 ). A second less detailed approach is represented by cellular automata, where
cars move between fixed-sized cells, e.g., Nagel and Rickert ( 2001 ). Interactions
between cars in neighboring cells are present, such that travel speeds and densities
can be modeled as disaggregated. Less details are present in mesoscopic models,
where detailed traffic demand is present however only aggregated supply, e.g., Ben-
Akiva et al. ( 1998 ). Macroscopic models represent the highest abstraction level,
where supply and demand is modeled on an aggregated level, e.g., Cayford et al.
( 1997 ). For a more detailed literature review related to traffic simulations, see
Charypar et al. ( 2007a ).
There is a trade-off between computation time and model detail. As MATSim
aims for large-scale simulation, it uses a queue-based approach, which in terms
of detail is located somewhere between the cellular automata and mesoscopic
approach. While the details in terms of implementation differ slightly, in general
all of MATSim's traffic simulators consider roads as active elements, which move
around cars. Each road link contains a queue which stores the entry time of each car.
Adjacent links collaborate with each other to assure that link capacity, free speed
travel time, intersection precedence, and space availability are taken into account
during the simulation. There are several implementations of the traffic simulation
available, which are presented in the following section.

9.2.3 QueueSim and JQueueSim

The first micro-simulation developed for MATSim is called QueueSim and is based
on a fixed-increment time advance model (Raney et al. 2003 ). In this model, vehicles
are moved along links in fixed time steps of 1 s. Although the model is quite flexible,
for larger simulations it is too slow because of the fixed simulation time step. A
parallel version of QueueSim was implemented leading to a significant speedup
(Cetin 2005 ). Both of these simulations are implemented in the C CC programming
language. In order to improve the maintainability of the code, MATSim was later
re-implemented in Java, including the nonparallel version of QueueSim, which
is called JQueueSim here. In recent years, the performance of JQueueSim has
improved, but the underlying simulation method remains the same.