Abstract:

Combining with the nonlinear and non-Gaussian characteristics of the traffic flow, we propose a traffic flow congestion event reconstruction framework based on the Sequential Monte Carlo (SMC) to deal with the congestion reconstruction problem. The simulation's states can get close to the real scene continuously when the data assimilation model assimilates the real-time sensor data constantly. The congestion event in real scene can be estimated based on the simulation data. Thus, the simulation model can simulate the congestion via different particle simulations and finally reconstruct the congestion event. Experimental results show that the framework can detect and reconstruct the congestion event on the real road network; the average error of the start position is 17m, while the average coverage rate of the congestion range is 82%.

Key words: SMC, DDDAS, traffic flow simulation, event reconstruction