Deep Reinforcement Learning–Based Control Strategy for Electro-Hydrostatic Active Suspension

A DRL (deep reinforcement learning) algorithm, DDPG (deep deterministic policy gradient), is proposed to address the problems of slow response speed and nonlinear feature of electro-hydrostatic actuator (EHA), a new type of actuation method for active suspension. The model-free RL (reinforcement learning) and the flexibility of optimizing general reward functions are combined with the ability of neural networks to deal with complex temporal problems through the introduction of a new framework called “actor-critic”. A EHA active suspension model is developed and incorporated into a 7-degrees-of-freedom dynamics model of the vehicle, with a reward function consisting of the vehicle dynamics parameters and the EHA pump–valve control signals. The simulation results show that the strategy proposed in this article can be highly adapted to the nonlinear hydraulic system. Compared with iLQR (iterative linear quadratic regulator), DDPG controller exhibits better control performance, achieves the EHA control objective at faster speed, and notably improves the ride comfort and handling stability of the car. Moreover, DDPG’s optimized valve–pump joint control strategy can reduce the energy consumption of the EHA system and improve the life of the hydraulic components while ensuring the control accuracy, solving the problem of low reliability of the active suspension system.