Research on online closed-loop drive control of vibration MEMS gyroscope based on the deep deterministic policy gradient algorithm

Vibrating tuning fork MEMS gyroscopes feature simple structures and ease of fabrication, making them one of the most widely used MEMS gyroscope devices. The sensing accuracy of such gyroscopes is strongly influenced by their drive state, and conventional closed-loop drive systems are typically designed using traditional PID controllers. However, due to the time-varying nature of MEMS gyroscope parameters under external disturbances, the fixed-parameter PID approach has limited effectiveness. To address this issue, this paper proposes an online closed-loop control method based on the Deep Deterministic Policy Gradient (DDPG) reinforcement learning algorithm. The controller is designed to adaptively adjust its control parameters in real time according to the current operating state, thereby accommodating environmental changes. A dynamic model of the MEMS gyroscope is developed in the Simulink environment, incorporating simulated temperature variations and external vibration disturbances. The intelligent agent observes the feedback signals to determine the current state and outputs appropriate control signals. Simulation results demonstrate that the trained agent can effectively stabilize gyroscope oscillations. Compared with the PID controller, the proposed method reduces the overshoot of the drive response signal by 90% and shortens the settling time by 50%. Under random vibration noise, the variance of the response signal is reduced by approximately 35%.