Tuning microscopic online simulations of freeway traffic with stationary detector data

Abstract

This paper discusses the continuous parameter calibration of micorscopic traffic models in online simulation. Al algorithm is described which uses a recent development in filter theory, the Unscented Kalman Filter. The algorithm is supplied with data of loop detectors and, of course, with the model. It adapts continuously the model parameters to the incoming data inorder to keep the online simulation as realistic as possible. the parameters to be filtered have to be chosen carefully, considering their probability density function, which should be roughly Gaussian-distributed. Numerical experiments in simple non-real scenarios was carried out using the "SK-Model" proposed by Stefan Krauss. The results show the ability ot the filter algorithm to calibrate at least some of the model parameters online.