Research And Realization On PF Based Traffic Flow Event Reconstruction

The rapid development of intelligent transportation system has an increasing demand to grasp real time traffic events on road network. But the former research about traffic flow reconstruction mainly concentrated on traffic data such as vehicle density, vehicle speed et al, and lack of exploration on event reconstruction like traffic congestion. Traffic flow has the characteristics of nonlinear, non-Gauss and high dimensional random. While particle filter, which is a sequential Monte Carlo algorithm and not limited by noise distribution, is widely used in nonlinear and non-Gauss systems. However, the traditional particle filter has the problem of particle degradation, and the resampling solution may also lead to particle enervation and high computational complexity, what's more its application in high dimensional random scene usually lack of accuracy and time efficiency. In this paper, we extended the research of Georgia State University SIMS laboratory on modeling and simulation of complex system, and realized traffic event detection and reconstruction in complex road scene by optimizing particle filter. The main contents of the paper are as follows.Firstly, the dynamic data driven traffic flow event reconstruction framework is built; like wise, the interaction interface and key techniques are determined. The simulation states can get close to the real scene continuously along with the data assimilation model assimilates real-time traffic data constantly. The congestion event can be detected by analyzing simulation results, and then it is reconstructed by multi-particle simulation. Secondly, a bidirectional particle filter algorithm(B3DPF) for high dimensional random scene is proposed based on the theory of generalized dynamic data driven system. The algorithm steps and key parameters can be adjusted according to the simulation results dynamically; simultaneously, the injection strategy of sensor data can be improved. It is verified that the implementation strategy and dynamic mechanism in B3 DPF can alleviate particle degradation and keep the particles' diversity; meanwhile, it has obvious accuracy and time efficiency advantages when dealing with high dimensional random scene. Finally, the traffic flow event reconstruction framework is realized by adding multi-thread mechanism, simulation state analysis and control, data storage and analysis, event detection and simulation, data assimilation interface to Mov Sim. The feasibility of dynamic execution strategy and information interaction is verified according to a simple linear road experiment; the experiment of the road network from Ming Imperial Palace to Zhongshan Gate in Nanjing proved that, the proposed traffic flow event reconstruction framework can detect and reconstruct the congestion event accurately.