Reinforcement Learning Based Adaptive LADRC for Active Magnetic Field Compensation in Magnetocardiography Systems

To suppress external magnetic field disturbances during Magnetocardiography (MCG) signal acquisition, this paper presents an active magnetic field compensation (AMC) method based on reinforcement learning based linear active disturbance rejection control (RL²ADRC). The proposed approach begins with precise parametric modeling of the AMC system, followed by the design of a linear active disturbance rejection control (LADRC) controller based on the identified system parameters. To improve the controller's disturbance rejection performance against magnetic disturbances with varying frequency characteristics, the LADRC parameters are initially tuned using the bandwidth-based method. A Soft Actor-Critic agent is then employed to adaptively adjust these parameters online. Simulation results demonstrate the effectiveness of the proposed RL²ADRC method in attenuating magnetic disturbances at different frequencies, thereby enabling high signal-to-noise ratio MCG measurements.