Comparing the Efficiency of Traffic Simulations Using Cellular Automata

The shift toward electric vehicles requires the development of an extensive public electric charging infrastructure. With the aim of simulating hundreds of configurations for charging stations, street directions, crossing, etc., we need to find the best solution in short periods of time to predict and prevent traffic congestion. Thus, we study different models to discretize and manage vehicle movements using a synchronous cellular automata, with an emphasis in reducing the amount of (frequently accessed) memory and execution time, and improving the thread parallelism. This is guided by the classical lemma of computer architecture “make the common case fast", thus optimizing those code sections where most of the execution time is spent. Experiments carried out for microscopic traffic simulations indicate that compiled languages increase run-time efficiency by more than 70(\times ). Then several strategies are studied, such as storing future velocities of each vehicle so that neighbor vehicles can benefit from this information. Using a single 12-core PC, we get to a total run-time for a unidimensional simulation that is very close to that reached by supercomputers composed of thousands of cores that use interpreted languages. This may also greatly reduce the energy consumed. Although some performance degradation may occur when complex situations are introduced (crossroads, traffic lights, etc.), this degradation would not be significant if the length of the streets were large enough.