The increasing complexity of simulation models raises difficulties related to their appropriate use and in particular their calibration and validation. Most applications involve a fair number of parameters. In many cases dynamic Origin-to-Destination (OD) flows are an important input, which is not directly observed and so have to be estimated jointly with the parameters of the various models. The available data is often partial and has substantial measurement errors. Most traffic simulation models are stochastic (Monte-Carlo). Therefore, their outputs are random variables. This has to be recognized by the optimization methods used in calibration and OD estimation methods that rely on simulation outputs.
We have been working on development of algorithms for efficient calibration of the model parameters and OD estimation of traffic simulation models.