O-D Matrix Estimation and Incident Detection in Urban Areas Using Artificial Neural Networks

The paper discusses the application of Neural Networks to OD estimation and assignment problems. Error Back Propagation Neural Networks, when suitably constrained to respect the road network structure, show an interesting formal analogy with both these problems. It is expected that the constrained neural network, because of its own structure, would be able to reproduce feasible states of the road network that are different to those learned from the observed sample. Following this analogy, a procedure is presented to detect both the occurrence and the extent of incidents in urban road networks. The procedure is then tested, as well as assignement and OD estimation problems, in a first experiment that confirms the goodness of this approach.