Number of Vehicles and Travel Time Estimation on Urban Traffic Network using Bayesian Network Model and Particle Filtering Method

Abstract

Travel time is one of the key variables that reflect the performance of traffic systems. The travel time is affected by the interaction between traffic demand (the number of incoming vehicles) and the characteristics of traffic supply (such as road capacity, traffic signaling, and driving speed). Therefore, any traffic conditions will result in different travel times. The traffic conditions are determined by the complex interactions of drivers, vehicles, and road or site characteristics. In this paper, the dynamics of traffic are modeled by taking a hydrodynamic theory approach, using standard assumptions commonly used in traffic engineering. Stochastic models of traffic evolution are derived and parameterized by turning ratio and the number of outgoing vehicles of each link. In addition, the travel time variability is modeled by statistical approach. The delay time experienced by a vehicle and its free flow speed are the two main sources of uncertainty that can be captured from the statistical model. The relationship between traffic dynamics and travel time is represented by the Dynamic Bayesian Network model. Using floating data (the position of the vehicle at each time interval) obtained from the Vissim simulator and probabilistic modeling framework, the paper focuses on a method for estimating travel time and traffic state. The particle filter method is used to estimate the traffic state of each link on the network, which is the hidden variable of the Bayesian model. The traffic process built in the Vissim simulator has been validated with the observed data collected directly on the real location of this study. Implementation of the estimation algorithm with Bayesian Network model has proved that this methodology is successful enough to give estimation results which are close to the actual data.