Intelligent Control Strategy of Vehicle Active Suspension Based on Deep Reinforcement Learning

• Main Authors: Zhu, Xuebin, Chen, Zhaoqun, Zhang, Sheng, Zhang, Cirun
• Format: Conference Proceeding
• Language: English
• Subjects: Deep learning; Deep reinforcement learning; Intelligent control; Reinforcement learning; Roads; Simulation; Suspensions (mechanical systems); TD3; Uncertainty; Vehicle active suspension; Vibrations
• ISSN: 2688-0938
• DOI: 10.1109/CAC57257.2022.10054782

Aiming at the uncertainty and nonlinearity of vehicle active suspension systems, this paper proposes a suspension vibration control strategy based on the twin delayed deep deterministic policy gradient (TD3) algorithm. A quarter-vehicle active suspension model and the time domain model of road excitation are established to develop the controller. Simulation results of vibration characteristics show that the intelligent control strategy has excellent performance and self-learning ability. Compared with the linear quadratic gaussian, proportional-integral-derivative, and sky-hook strategy, the deep reinforcement learning controller has better robustness and generalization ability, which can adapt to sophisticated driving conditions of the vehicle, and has an extensive application prospect in the development of the active control arithmetic for suspension.