Towards congestion detection in transportation networks using GPS data Armando Bazzani, Bruno Giorgini, Riccardo Gallotti, Luca Giovannini, Monica Marchioni and Sandro Rambaldi Dipartimento di Fisica Via Irnerio 46 40126 Bologna Italy www.centrogalvani.unibo.it www.physycom.unibo.it Abstract—The future Information Communication Technolo- gies will allow the collection of a large amount of data on individual mobility. The data analysis will not only provide information on the human mobility peculiarities, but will also open new possibilities for studying the criticalities of a complex system. In Italy a sample of 2% of the vehicle population is monitored by a GPS system for insurance reason. We have analyzed the data in the metropolitan area of Rome for a whole month (May 2010) to reconstruct an empirical Fundamental Diagram based on the real time reconstruction of the speed, density and flux on the whole road network. The main difficulties are the low sampling rate of GPS data set and the heterogeneity of the road network in the metropolitan area of Rome and of the drivers’ behavior. We present the first results of this research discussing the future perspectives for prevention and governance of traffic congestions. I. I NTRODUCTION Traffic congestion in modern metropolis is strictly linked to life quality and human freedom, especially in the modern age and in the actual globalization context[1]. The Information Communication Technologies (ICT) development offers new possibilities for detecting microscopic data on human mobility. It is not improper to say that the transportation technologies and the means of transport are structurally connected with the ICT. More precisely we are in presence of two intertwined but independent processes: the generalizing of mass mobility, and the diffusion of tools allowing a rapid exchange of large bulks of information (ICT)[6]. In the next future individuals could get information on safety, on possible congestions, on different mobility paths and means via smart mobile phones, portable computers and so on, coupling information, so that it becomes possible for the single to self-organize his own individual mobility[2]. But a theoretical effort is needed to define suitable macroscopic observable to measure the congestion degree from the empirical observations. In the past the engineers has introduced the Fundamental Diagram to study the traffic flow dependence on the average vehicle density in a highway[8]. Experimental measures show the existence of a critical density, corresponding to the maximal flow, at which a phase transition occurs due to the appearance of the stop and go dynamics. The critical density depends both on the microscopic dynamics of vehicles and on the geometrical features of the road. If the density overcomes the critical value the FD is not a continuous single value function and both experimental data and numerical simulations show a fuzzy behavior for the flow values. The existence of a FD for a whole road network, where the effect of road crosses is relevant, has been proposed in previous works by using theoretical models [3] and empirical observations [4]. The FD provides macroscopic information on the congestion state of the network. Indeed the FD can be interpreted as the convolution of the FDs for each road. Generally one still expect the existence of a critical average density for the optimal traffic flow on the network, but in this case the FD curve is continue even for density greater than the critical one. However for such density value the network cannot be in a stationary state due to the congestion spreading phenomenon due to the increase of local queues, which reduce the average total flux. Therefore the FD is a powerful tool to represent the macroscopic traffic state and to develop a now- casting approach to the traffic dynamics on a urban road net- work (i.e. a forecasting of the traffic state within the next hour). In this paper we illustrate some results on fluxes and velocities measures on the whole metropolitan area of Rome using GPS data, which sample individual trajectories on a ≃ 2% of the vehicle population. Despite the low spatial resolution for the individual trip, we have developed algorithms[5] that allow the trajectory reconstruction and an estimate of the flux on each road in a short time scale. To understand the assumptions and problems at the base of a FD computation for a whole network, we perform numerical simulations using an optimal velocity model in a simple, but realistic road network. A preliminary analysis of the empirical FD using GPS data points out the relevance of non-homogeneity effects both in the roads and in the vehicle dynamics, which introduce a great variability in the velocity-flux relation at local level. However our analysis defines a road map towards a real time reconstruction of the FD taking advantage of the future ICT. II. THE GPS DATA SET As case study we analyze the whole metropolitan area of Rome, corresponding to a region of 20 x 20 km along the entire month of May 2010. The recorded GPS data for individual trip have a spatial resolution of 2 km or a time resolution of 30 seconds. Moreover, one datum is always recorded when the vehicle engine starts or stops. Each datum 2011 IEEE International Conference on Privacy, Security, Risk, and Trust, and IEEE International Conference on Social Computing 978-0-7695-4578-3/11 $26.00 © 2011 IEEE DOI 1455