Advanced queuing traffic model for accurate congestion forecasting and management

Traffic congestion can be explained as a percolation in a dynamical system on a network structure because of the existence of a maximum flow rate, a finite transport capacity at crossing points and finite road capacity. In this thesis a mesoscopic traffic model is developed to study the congestion formation on a road network. The goal is to propose a predictive observable for the rise of congestion based on the traffic load fluctuations using the flow-density fundamental diagram of the network. The model simulates the vehicle dynamics using an average optimal velocity model along the road and introduces finite flow rates at the crossing points, also considering the traffic lights dynamics. Simulations are performed on a Manhattan-like network where all junctions are traffic lights or roundabouts. The main features of the congestion transition have been studied, both in stationary and transient states. A statistical analysis of traffic flow and density fluctuations was performed to highlight their predictive properties for congestion detection. From such simulations, flow fluctuations and normalized density fluctuations are highlighted to predict the rise of congestion when the system is in a stationary state. Furthermore, an optimization algorithm based on vehicle density is proposed and tested on a network consisting in traffic lights only. This algorithm shows good results regarding density fluctuations, which result lower with respect to unoptimized simulation for the same load. Despite that, the algorithm performs poorly on traffic flow, reducing it for the same load with respect to unoptimized simulations, mainly at high density. Thus, it appears unsuitable to manage congested systems. Finally, a test using data following a real trend is done on Via Stalingrado in Bologna, confirming the goodness of the model.