Stochastic Lagrangian Traffic flow modeling and real-time traffic prediction

摘要

Lagrangian Traffic model which follows a platoon of vehicle has the benefit of convenient utilization of individual vehicle data and easy distribution of traffic information to the drivers. The objective of this paper is to develop a stochastic Lagrangian traffic flow model which uses probe vehicle data to predict traffic status. The proposed probing method tracks vehicles in pairs to collect speed and spacing between vehicles. The traffic flow model utilizes unscented Kalman filter (UKF) with dual estimation to update model parameters and estimated current traffic in real-time. The proposed model was validated by empirical highway traffic data, and the result showed that the stochastic model has an overall 20% improvement in estimating current traffic state comparing to the estimation from deterministic model. The predictive ability of the model with an average of 15% error for 3-sec prediction can be used to compensate for the latency of data processing in real-time application. This paper also demonstrated a new method to predict the unexpected jam traffic phase using estimated model parameter with an average lead time of 6.76 sec, which allows drivers to be prepared for potential stop-and-go traffic.