The maglev train, running at high speed, needs sensorless control to estimate the speed and position based on calculating the pole phase angle of the long-stator linear synchronous motor (LSM). This article proposes a novel sensorless control strategy using extended full-order state adaptive observer. The nonlinear model of LSM is established, taking into account the variation of vertical and longitudinal freedom parameters. Then, an extended full-order state adaptive positon observer is designed on the fusional basis of full-order state observer and extended state observer, which settles the delay problem of non-ideal control element at high speed. Furthermore, optimizing speed regulator to achieve efficient decoupling control of dynamicity and stability for the problem of tracking delay and load disturbance, through analyzing the influence of feedback gain, observer adaptive parameter, speed regulator parameter adjustment on the sensorless control system. The average estimation error is limited to 0.2rad at maximum speed. And the control effect is improved. Theoretical analysis is presented, and the proposed approaches are verified by experiments on Rt-lab, a semi-physical simulation platform.