Fast algorithms for the approximation of a traffic flow model on networks (Q2496589)

The main purpose of the paper is to introduce and examine new computation algorithms for the fluid-dynamic mathematical modelling of the flows on networks, particularly traffic and telecommunication networks. The numerical approximation of the traffic models based on the conservation equation (for cars) is devised with simulation algorithms involving (classical) numerical Godunov (G) scheme and 3-velocities Kinetic (K3V) schemes of first and second order. The main contributions of the paper consist in developing two new algorithms to compute the solutions for traffic flow on a network for a piecewise constant flux function: Fast Godunov (FG), and Fast Shock Fitting (FSF) schemes. First, the performance of classical Godunov computation scheme is improved by substituting the usual computations for numerical flux with some ``if then'' instructions based on a priori given flux values. The obtained FG scheme is shown to significantly speeding up the computations. The second new numerical approach, FSF, is based on the existence of a single separating shock, with two phases, one phase for light traffic and another phase for heavy traffic. The FSF algorithm constructs solutions by localizing shocks, exploiting the existence of a unique separating shock, tracking it with an exact fitting procedure, and computing the solution by a shift procedure. To notice that FG scheme algorithm is applicable for general initial configurations or in the case of conservation laws with a source term, while the FSF scheme algorithm cannot be used for this kind of problems. The K3V scheme of second order is also applied to the same flux function in order to compare the computations with the newly proposed algorithms. Numerical experiments show the performance of the schemes regarding the accuracy and execution time. The FG algorithm is proved to save more than 50\% of CPU time of computation, in comparison to the classical G scheme. On the other hand, the FSF algorithm is shown to outperform the FG one, with an increasing percentage (around 70\%) for large networks and space discretization steps. These results are of special importance when applied to the road networks of great cities, with a large infrastructure of traffic networks.