Combining simulative and statistical approach for short time flow forecasting

In this work we propose an approach based both on Dynamic Traffic Assignment (DTA) model and Kalman filtering for estimating short time forecasts of time-dependent link flows in a freeway, using available data on link traffic volumes. In particular, a sequential procedure using a Generalised Least Squares (GLS) approach is adopted to estimate the time-dependent trip tables. After the estimation of origin-destination matrices, a DTA procedure, based on an enhancement of a mesoscopic model developed by one of the authors, is performed in order to generate time-dependent link flow contribution factors (that are a dynamic generalization of the path-link incidence matrix for the static case) making up the so called assignment matrix. The obtained assignment matrix is then used within a Kalman filtering procedure, to directly estimate short time forecasts of traffic flows. Differently as done in some approaches recently presented in literature, in the proposed one Kalman filtering is not used to dynamically estimate the values (or the biases) of the OD matrix but to directly estimate link flows and the estimation of the assignment matrix used in the state equation of the Kalman filter is performed at each time interval in which simulation is conducted. A set of off-line experimentation have been also conducted comparing forecasts obtained both with the presented method and through a simple simulation. Two consecutive weeks flow counts at barriers of a close system freeway located in southern Italy has been used in order to validate the proposed procedure. In particular computational results and performances, in terms of time consumption and reliability of forecasts, are presented.