A Framework to Characterise, Estimate, and Predict Vehicle Class-Agnostic Traffic States and Class-Wise Speeds for Mixed Traffic Conditions

• Published in: IEEE access Vol. 12; pp. 106211 - 106235
• Main Authors: Ashok, Abirami, Rama Chilukuri, Bhargava
• Format: Journal Article
• Language: English
• Published: Piscataway IEEE 2024; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: class-wise speeds; Data models; Driving conditions; Flow-density-speed relationships; Kalman filters; Kinematics; Mixed traffic; Parametric statistics; Prediction models; Predictive models; Probes; Roads; speed-based state characterisation; Traffic; Traffic control; Traffic management; Traffic speed; traffic state prediction; Travel time; travel-time data; Vehicle dynamics; Vehicles
• ISSN: 2169-3536; 2169-3536
• DOI: 10.1109/ACCESS.2024.3437172

For homogeneous traffic, where all vehicles are the same type, the traffic state is characterised by speed, flow, density, queue length, etc. In mixed traffic conditions, variations in static and kinematic characteristics among vehicles and the resulting asymmetric interactions that arise, these state variables are inadequate to represent class-wise behaviours. This paper proposes a novel framework for characterising mixed traffic conditions based on vehicle class-wise speeds rather than a single value of the aggregated stream speed. Also, it proposes an area occupancy-based approach to estimate class-wise speeds from class-agnostic disaggregated travel-time data. The empirical validation of the proposed traffic state definitions demonstrates their generalisability. Finally, parametric and non-parametric prediction models are also developed for state and class-wise speed predictions. The empirical results demonstrate that the joint prediction approach (simultaneous prediction of multiple classes using the proposed state definition) is more accurate, computationally effective, and more efficient for practical applications than the marginal predictions using class-wise speed predictions. Moreover, the order of the class-wise speeds is more robustly preserved in the former than in the latter. This research can open doors for a new family of class-wise speed-based traffic management strategies and applications for mixed traffic conditions.