This content is not included in your SAE MOBILUS subscription, or you are not logged in.

Multi-objective Optimization for Connected and Automated Vehicles Using Machine Learning and Model Predictive Control

Journal Article
14-11-02-0014
ISSN: 2691-3747, e-ISSN: 2691-3755
Published November 05, 2021 by SAE International in United States
Multi-objective Optimization for Connected and Automated Vehicles
                    Using Machine Learning and Model Predictive Control
Sector:
Citation: Zhu, H., Song, Z., Zhuang, W., Hofmann, H. et al., "Multi-objective Optimization for Connected and Automated Vehicles Using Machine Learning and Model Predictive Control," SAE Int. J. Elec. Veh. 11(2):177-187, 2022, https://doi.org/10.4271/14-11-02-0014.
Language: English

Abstract:

Connected and automated vehicles have attracted more and more attention, given the benefits in safety and efficiency. This research proposes a novel model predictive control method in order to improve energy efficiency and ensure a safe spacing between vehicles. The proposed algorithm focuses on mixed traffic flow, which is more realistic than one that only includes autonomous vehicles. A high-fidelity energy loss model of an electric vehicle is adopted to improve the control’s performance. A data-driven car-following model using machine learning is integrated in the framework of model predictive control to predict the behavior of human-driven vehicles. Its effectiveness in increasing energy efficiency is validated under two driving cycles. In the case of the scaled urban dynamometer driving schedule, the energy loss and the maximum spacing between the autonomous vehicle and the human-driven vehicle decreases by 6% and 18%, respectively, when compared with the baseline model predictive control without the consideration of interaction between the autonomous vehicle and the human-driven vehicle. In the scenario of the scaled city driving cycle, the energy loss of the autonomous vehicles also reduces by 3%, while the maximum and average spacing does not change significantly. The sensitivity of the optimization results to several parameters of the energy loss model is finally analyzed, and the robustness of the proposed algorithm is validated.