Research on Refined Modeling and Control Strategies for Electro-Hydrostatic Active Suspension Systems

To address the challenges of strong nonlinearity, time-varying parameters, and poor adaptability of traditional control methods in Electro-Hydrostatic Actuator (EHA)-based active suspension systems, this paper proposes an end-to-end control strategy using the Deep Deterministic Policy Gradient (DDPG) algorithm. First, a high-fidelity dynamic model of the EHA hydraulic system is established, incorporating core components such as the PMSM, fixed-displacement vane pump, double-acting hydraulic cylinder, and servo valve. This model accurately captures the multi-physics coupling characteristics of the EHA by integrating motor electromagnetic behavior, nonlinear hydraulic pressure-flow relationships, and mechanical load dynamics. Second, to overcome controller design difficulties caused by system nonlinearity, an end-to-end DDPG-based control framework is developed. The framework uses suspension vertical acceleration, suspension dynamic deflection as state observations, while generating control outputs for EHA. By directly optimizing the control policy through deep reinforcement learning, the method avoids the linearization simplifications and parameter sensitivity issues inherent in traditional modeldependent controllers. Simulation results demonstrate that, the proposed method reduces the root mean square (RMS) value of vertical body acceleration by 42.08 %, which validates the precision of the approach in handling complex nonlinear systems.