Description
Back to topNowadays, a huge amount of traffic data from different sources (magnetic loop detectors, video cameras, radars, floating car data, bluetooth, etc) is available, which can be used to calibrate and supplement mathematical models, providing mixed model- and data-driven modeling approaches for accurate simulation of road traffic in real-life transportation networks, with applications in real-time decision support systems and urban planning. Indeed, the heterogeneity of traffic conditions in congested regimes makes it hard to obtain a good matching between simulations and reality, thus preventing from getting reliable traffic state predictions beyond short time horizons (5-10 min). Even if enhanced models accounting for heterogeneities and multi-scale factors have been developed over the years, their calibration is even more challenging, requiring high quality data. On the other hand, raw data sets often contain corrupted or missing values, which makes the direct information too poor to be used as such. In this perspective, statistical learning (such as Gaussian Processes and Neural Networks) and mathematical modeling can complete each other to produce a more accurate and complete description of the traffic dynamics on the considered road network.
The aim of this workshop is then twofold: on the one hand, we are interested in analyzing information derived from traffic data using innovative machine learning methods and at exploiting them within deterministic PDE models. On the other hand, the coupling of probabilistic methods with physically grounded mathematical models is expected to ensure traffic predictions remain plausible in regimes with no or corrupted data.