In the framework of the stadtinfokoeln project, the DLR (MS) is together with the ISB (RWTH Aachen) responsible for the prediction of future traffic states. Here we discuss and compare two different techniques for the forecast of traffic counts at specified detectors: a so-called fitted template technique (FTT) that fits a general detector pattern to the current traffic state and a method using a neural network (multi-layer perception, MLP). Both techniques reproduce nicely the overall evolution, the MLP at a slightly higher precision, however at the price of being numerically less efficient.
Abstract