Optimal dynamic semi-active control of high-speed elevator car horizontal vibration based on DDPG

When the elevator car operates at high speed, it can experience unpredictable horizontal vibrations due to factors such as guide rail unevenness, guide shoe nonlinearity, and variations in load, all of which affect both safety and ride comfort. To effectively suppress the horizontal vibration of the car, this paper proposes an optimal dynamic semi-active control method based on deep deterministic policy gradient (DDPG). First, a dynamic model of the elevator car’s horizontal vibration system is established, accounting for the nonlinear coupling between the guide rail, guide shoe, car frame, and car body, while the unevenness of the guide rail is simulated using power spectral density. Second, a Markov decision process model is formulated for controlling horizontal vibrations, with the state space, action space, and reward function designed for deep reinforcement learning. Third, to improve the generalization performance, the DDPG algorithm is enhanced by normalizing the state space of the elevator, and the controller learns the optimal dynamic semi-active control strategy by interacting with the simulation environment, and achieves real-time control of the car’s horizontal vibration. Simulation results demonstrate that, compared to passive control, Mixed SH-ADD control, PID control, and DQN-based control, the DDPG-based method reduces the average horizontal acceleration of the elevator car by 68.90%, 45.46%, 21.35%, and 20.95%, respectively, effectively mitigating horizontal vibration.