To solve the problem of control performance degradation caused by unknown interference in maglev system, an active disturbance rejection control method (adaptive linear active disturbance rejection control,ALADRC) based on gradient information adaptive observation bandwidth is proposed. Firstly, a nonlinear model of single-point suspension system is established, and the stability region of auto-disturbance rejection parameter is deduced theoretically, and the concept of critical bandwidth is obtained. Secondly, the iterative formula of the adaptive linear extended observer is derived based on the minimization of the observation error, which enhances the stability of the system parameters. Even if the current bandwidth may diverge the system, ALADRC can automatically adjust to the relatively optimal stable bandwidth point, making self-tuning feasible. At the same time, when encountering disturbances, the bandwidth will be adjusted accordingly to enhance the immunity of the system. Then, the convergence of the observed bandwidth of ALADRC under different learning rates is simulated. It is concluded that higher learning rate leads to faster convergence of the observed bandwidth, and the final convergence bandwidth value is relatively larger. In addition, the order of magnitude of the learning rate can be reversely derived according to the critical bandwidth and the system unit scale to simplify the adjustment of the learning rate. Finally, the control performance of PID(proportional integral derivative), LADRC(linear active disturbance rejection control) and ALADRC are compared respectively on a single-point suspension platform. The experimental results show that, compared with PID and LADRC, ALADRC has the best comprehensive control performance, which can realize fast non-overshoot flotation with excellent self-regulation and disturbance immunity.