Using Meta Heuristic Algorithms to Improve Traffic Simulation
Publication Type
Original research

Simulation today is one of the most used tools in science and engineering. Traffic engineering is no exception. Simulators to be usable passes through processes of verification, validation and calibration. All simulators are based on assumptions and parameters that need to be calibrated so as to be practical in real world applications. Some parameters change from site to site. Therefore, the calibration process is often needed. Calibration can be seen as an optimization process that seeks to minimize the difference between observed and simulated measures. The question of which optimization technique suits more for this particular problem remains open. In this paper the convergence velocity of main heuristic optimization techniques, namely Genetic Algorithm (GA), Tabu Search (TS), Particle Swarm Optimization (PS) and Simultaneous Perturbation for Stochastic Approximation algorithm (SPSA) were used to calibrate a traffic simulation model called SUMO. The results of the calibration of the mentioned optimization techniques were compared. Classical optimization techniques, namely Neldear-Mead and COBYLA were used as a baseline comparison. Each technique has its own parameters that affect convergence velocity. Therefore, optimization techniques themselves need to be calibrated. However, TS and PS are not widely used to calibrate traffic simulators. They perform well in this particular problem. PS is highly parallel compared to the TS and SPSA. The paper shows that classical optimization techniques are not suitable for this particular problem, PS and TS appear to be better than GA and SPSA. PS seems to be a promising optimization technique.

Journal of Algorithms and Optimization, Vol. 2 Iss. 4, PP. 110-128
Publisher Country
Publication Type
Online only