Particle Filter Based Traffic Data Assimilation with Sensor Informed Proposal Distribution

摘要

This paper presents a particle filter with sensor informed proposal distribution for traffic state estimation. An agent-based traffic simulator is employed to simulate the traffic network and vehicle behaviors with the help of the sensor and accident component. The proposed framework estimates the average speed of a traffic network along with the location of any traffic accidents. This framework uses observed data to construct the system state by proposing local sensor proposal distributions. The performance is validated and evaluated by means of identical twin experiments carried out on both Bootstrap and Sensor Informed Particle Filter (SIPF). The result shows that the proposed data assimilation framework outperforms Boot-strap Particle Filter with respect to accuracy and efficiency.