Section: Scientific Foundations

Calibration and convergence of integrated models

When implementing such integrated models, one of the main difficulties lies in the calibration process. Feedback from past experience tend to show that this task is very difficult to proceed, and needs an important experimental expertise. In practice, the equilibrium models such as Tranus converge with difficulty, and the current algorithms do not enable to easily correct the parameters to obtain the convergence. This convergence is a key-point of the calibration: the goal is to be able to reproduce a reference state in a stable manner.

Calibrating such models involve the estimation of a large number of parameters, that are difficult to estimate from the data. In general, and this is true for the case of Tranus, these parameters are currently adjusted by hand, through a long process of trials / errors. The calibration can typically take up to 6 months for a medium size model (about 100 geographic zones, about 10 sectors including economic sectors, population categories, employment categories). So far, ways to optimize these parameters in an automatic or semi-automatic manner do not exist, and it is not possible to guide the solution towards a good representation of the observed reality, but by hand. Here, it is not only to converge towards a stable state that matters, but also to make sure that this state corresponds to the chosen reference of the urban system we aim at mimicking.This is of course a crucial condition to ensure the prospective results produced afterwards by the models to be relevant.

Finally, let us note that knowledge of uncertainties has to to be taken into account in the calibration process to obtain robust and reliable results. This is detailed in the following section.