Abstract: To address the issues of poor adaptability and complex parameter tuning in active suspension control strategies, this paper proposes an active suspension control strategy based on the reinforcement learning soft actor-critic (SAC) algorithm. A dynamic model of the suspension system is established, based on which a policy neural network is constructed. A state-adaptive temperature parameter update mechanism is designed to improve convergence efficiency and control accuracy. Simulation results demonstrate that the proposed control strategy exhibits excellent robustness and adaptability. Compared with passive suspension, proportional-integral-derivative (PID) controlled active suspension, and skyhook algorithm-based semi-active suspension, the ride comfort index is improved by 8.1% to 22.6%, providing an effective solution for intelligent control of active suspensions.

Key words: deep reinforcement learning; active suspension control; SAC algorithm; vehicle ride comfort; control strategy robustness

摘要： 针对汽车主动悬架控制策略适应性差、参数调节复杂的问题，文章提出一种基于强化 学习软动作-评价（SAC）算法的主动悬架控制策略。在建立悬架动力学模型的基础上，构建 策略神经网络，并设计状态自适应的温度参数更新机制，以提升收敛效率与控制精度。仿真 结果表明，该控制策略具有良好的鲁棒性与适应性，相较于被动悬架、比例-积分-微分（PID） 控制主动悬架及天棚算法半主动悬架，行驶平顺性指标提升了 8.1%～22.6%，为主动悬架智 能控制提供了一种有效解决方案。

关键词: 深度强化学习；主动悬架控制；SAC 算法；车辆平顺性；控制策略鲁棒性