Ride comfort improvement for maglev trains using active disturbance rejection control with an improved RBF neural network

Abstract

To suppress the vertical vibration of maglev trains induced by track irregularities and improve ride comfort, this paper proposes an improved radial basis function neural network-based active disturbance rejection control (IRBF-ADRC) method. The proposed approach employs a three-dimensional input RBF neural network incorporating the tracking error and its variation trend and introduces a momentum term to optimize the learning process. This enables the online adaptive tuning of the extended state observer (ESO) gains, thereby enhancing the controller’s capability to observe and compensate for complex disturbances. Based on the measured track irregularity data from the Shanghai maglev line, comparative verifications are conducted over a speed range of 100–400 km/h. The results demonstrate that, compared with the conventional PID control and the ADRC in reference (Wang et al., 2024), the proposed method reduces the root mean square (RMS) acceleration of the levitation electromagnet (mover) centroid by an average of 80% and 20%, respectively, while maintaining a safe levitation air gap margin. Moreover, within the human-sensitive frequency band of 4–12.5 Hz, the vibration energy is significantly suppressed without increasing excitation loss, achieving both ride comfort and energy efficiency.