Using Parallel Particle Swarm Optimization for Auto-Tuning of Traffic Micro-simulations in Heterogeneous Conditions

This study contributes towards enhancing two dimensions of traffic micro-simulations. Firstly, it provides a methodology to reduce the calibration time and effort of micro-simulation models by integrating knowledge of multithreading techniques and evolutionary algorithm. Secondly, it implements this methodology to calibrate a traffic micro-simulation with specific consideration of heterogeneity in traffic stream. The first contribution looks at the importance of calibration and its related difficulties. It applies the particle swarm optimization algorithm for auto calibration of traffic micro-simulations simplifying the process by eliminating human intervention during trial-and-error procedures. To shorten the execution time, parallelization of the algorithm was implemented using 32 CPUs in this study. The results implied that this method could reduce the running time by approximately 25 times when compared to unparalleled evolutionary algorithms. The second contribution deals with the different behavior of drivers in heterogeneous traffic conditions. This consideration could be very important, in particular when considering the increasing number of heavy vehicles on the road. The developed parallel particle swarm optimization algorithm was implemented as a case study to calibrate a micro-simulation model which specifically considers heavy vehicles and passenger cars and their interactions. The method was also used for calibration of the micro-simulation with the existing models. The results show that this approach could enhance the performance of micro-simulations in estimation of traffic measurements.