Real-Time Traffic Prediction Considering Lane Changing Maneuvers with Application to Eco-Driving Control of Electric Vehicles

Abstract

Emerging vehicle sensing and communication technologies allow for real-time information exchange between connected vehicles (CVs) and intelligent infrastructure. This presents a unique opportunity for predicting traffic states such as speed and density. A promising application of traffic prediction is eco-driving speed control of CVs, which requires future traffic information along the look-ahead time horizon. However, it is challenging to obtain accurate real-time traffic prediction for the next 10-15 s, particularly for mixed traffic involving both CVs and human-driven vehicles (HVs), complicated further by the presence of lane changing maneuvers. In this article, we address this pressing problem by integrating a macroscopic traffic flow model for prediction with a microscopic vehicle model for speed control. Specifically, we modify the well-known second-order Payne-Whitham (PW) model to account for the impacts of lane changing on traffic state evolution, based on which we develop a traffic prediction framework capable of handling mixed traffic. CVs provide partial measurements of traffic states, while the unknown states are estimated using an unscented Kalman filter (UKF). Consequently, future traffic states are obtained by propagating the PW model forward in time, and optimal eco-driving speed controls are obtained for electric vehicles (EVs) using the prediction results. The proposed approach is evaluated using ample traffic data collected from Simulation of Urban MObility (SUMO). The results show an average energy benefit of 6.6% for the ego vehicle considering all the simulated scenarios, among which the maximum energy benefit is about 16.18%.