Optimization of operational stability for ultra-high-speed elevators under time-delay disturbance: a study on DRL-GRU method

Abstract

Purpose To solve the poor real-time performance of active control systems for ultra-high-speed elevator horizontal vibration suppression, which is caused by sensor signal delays and unknown external disturbances. Design/methodology/approach First, a nonlinear 8-degree-of-freedom dynamic model of elevator horizontal vibration with a state delay mechanism was established. Then, a GRU-based state estimator was developed and a modified DDPG-NoisePER algorithm was proposed, integrated into GRU-DDPG-NoisePER to train the optimal delayed-state control strategy. Findings The GRU-DDPG-NoisePER-based strategy reduces the root mean square of car vibration acceleration by over 59%, outperforming DDPG-NoisePER by more than 1.02 times and DDPG-PER by more than 1.32 times. Practical implications The GRU-DDPG-NoisePER algorithm can be directly applied to ultra-high-speed elevator horizontal vibration control. It solves real-time issues caused by state delays, reduces equipment wear from excessive vibration, lowers maintenance costs, and provides a feasible technical solution for elevator manufacturers to optimize product performance in practical engineering scenarios. Social implications As a core transportation tool in high-rise buildings, stabilized ultra-high-speed elevators improve passengers' travel comfort (reducing vibration discomfort) and safety. The technology also drives technical upgrading of the elevator industry, aligns with smart city transportation needs and indirectly guarantees efficient and reliable urban vertical transportation. Originality/value This study innovatively integrates GRU with a modified DDPG (DDPG-NoisePER) to form GRU-DDPG-NoisePER. It effectively addresses state delay issues, improves DRL training efficiency, enhances vibration control effect and has strong practical application potential.