Adaptive weight distributed model predictive control for active suspension based on deep deterministic policy gradient

Abstract

To improve the ride comfort and attitude stability of the vehicle under complex driving conditions, this paper proposes a distributed model predictive control (DMPC) strategy with an adaptive weight-tuning mechanism based on the deep deterministic policy gradient (DDPG) algorithm for the active suspension system. The proposed method addresses the strong coupling among body vertical, pitch, and roll vibration-control objectives. It also reduces the reliance of conventional controllers on empirical parameter tuning and improves their adaptability to varying conditions. This study establishes a seven-degree-of-freedom full-vehicle active suspension model and decomposes it into a body subsystem and four wheel subsystems according to the coupling relationships. Then, a distributed predictive control framework is constructed. In this framework, local receding-horizon optimization and limited information exchange are used to achieve coordinated control. Furthermore, the DDPG algorithm learns the dynamic characteristics of the system online and adaptively adjusts the weighting parameters of the DMPC controller in real time. This enables dynamic allocation of control effort under varying operating conditions. The simulation results obtained from a high-fidelity CarSim co-simulation platform show that the proposed method effectively suppresses body vertical, pitch, and roll vibrations under different operating conditions. In addition, the proposed strategy reduces the average computation time compared with conventional MPC. Hardware-in-the-loop experiments further validate the effectiveness and real-time performance of the proposed controller.