A Gaussian Mixture Model and Data Fusion Approach for Urban Travel Time Forecast

Travel time information is gaining increasing importance as traffic performance measure from the perspective of both drivers to understand traffic conditions and network management to properly monitor and control the evolution of traffic conditions. Aiming at forecasting path travel times in congested urban areas, this paper proposes a data-driven approach in which traffic data are collected via Bluetooth and mobile Floating Car Data devices. Such data are used to improve the accuracy of the detected information by means of a Gaussian Mixture Model (GMM) and a Bayesian data fusion approach. The GMM is applied to estimate the travel time and not only its distribution and it is calibrated for each time interval and updated with every available data on real time. An Auto-Regressive Integrated Moving Average model is used for the travel time forecast. An application to a real-life test case in the city of Rome shows the goodness of the proposed approach for better online network performance forecast.