Research On Traffic Flow Forecasting And Optimization Based On Particle Filter

ABSTRACT:Traffic flow forecasting is a hot topic in transportation research field. And is also one of basis in effective management of traffic safety. Traffic system is a complex system composed of people, vehicles and roads, its operation will not be safe and orderly without the technology of traffic control and traffic guidance. However, traffic control and guidance are based on the forecasting. Therefore, forecasting the traffic flow accurately and timely could be able to reduce traffic congestion in a certain degree, it could keep the traffic state safe and smooth.Serveral years of research by scholars at home and abroad, they have established a variety of traffic flow models. But these methods mostly based on linear forecasting model. With the increase of transportation system's complexity, traditional forecasting methods are shown limited. For this situation, this paper attempts to introduce Particle Filter algorithm (PF). As a nonlinear forecasting method, PF can be used in state space models with any forms. In order to verify the superiority in nonlinear system's forecasting, the paper took Kalman Filter algorithm (KF) and PF for an example to do a simulation experiment pointing to the same nonlinear model, the result showed that PF had better forecasting performance in time-varying systems with nonlinear models than KF. This is also one of the reasons to choose PF as a forecasting method.When forecasting traffic flow of urban freeway, the paper selected a freeway section which belongs to north third ring road in Beijing, and collected information on traffic flow related. By analyzing the state of traffic flow characteristics to establish traffic flow model, and used PF to forecast traffic flow of the road. The result showed that PF could do a good forecasting, and had better applicability.However, in the course of the trial, with the increase of number of particles' iterations, only a few particles have greater weights, and majority of particles appears degradation. Pointing to this degradation, this paper tried to use the Genetic Algorithm (GA) to optimize resampling process of PF to get better particles. It is verified that the improved method had a better forecasting performance.