Calibration of Microsimulation with Heuristic Optimization Methods

Abstract

Model calibration is a crucial step in building a reliable microscopic traffic simulation application because it serves as the additional check to ensure that the model parameters accurately reflect the local driving environment so that decisions made on the basis of these results will not be misinformed decisions. Because of its stochastic nature and complexity, the calibration problem, usually formulated as an optimization problem, is often solved by using heuristic methods. To date, calibration is still a time-consuming task because many adopted methods require many simulation runs in search of an optimal solution. Moreover, many aspects of the calibration problem are not fully understood and need further investigation. In this study, another heuristics calibration algorithm is developed on the basis of the simultaneous perturbation stochastic approximation scheme and applied to calibration of several networks coded in Paramics. The results indicate that the new heuristic algorithm can reach the same level of accuracy with considerably fewer iterations and computer processor time than other heuristic algorithms such as genetic algorithms and the trial-and-error iterative adjustment algorithm. Applications of all three heuristic methods in a northern California network also reveal that some model parameters affect the simulation results more significantly than others. These findings can help modelers better choose calibration methods and fine-tune key parameters.