Abstract
When dealing with nonlinear time series of car traffic on highways, one of the outstanding problems to be solved is completion and continuation of data in space and time. To this end, the underlying process is decomposed into stochastic and deterministic components. The former is approximated by Gaussian white noise, while the latter refers, apart from always existing trends, to the space- and time-dependent jam propagation process. Jams are modelled in terms of dynamical Bayesian networks with radial basis functions involved. The models developed are used to tackle travel time estimation and prediction. Results are obtained for one of the most crowded traffic areas of Europe, namely the ring-like highway around Cologne.