Virtual coupling technology is currently a focal point within the railway industry. Through vehicle-to-vehicle communication, virtual coupling allows trains to form a unified train unit, thereby reducing inter-train spacing and enhancing transportation efficiency. The complexity of tracking operations and the shorter spacing between virtual coupled trains place high demands on controller performance. This paper designs an adaptive model predictive control (MPC), enabling automatic adjustment of control parameters with the train's operation. Firstly, this paper establishes a train dynamics model based on an improved single-point mass model. Secondly, considering the characteristics of train tracking operations, this paper formulates rules for varying control parameters, thus designing an adaptive MPC. Finally, to validate the effectiveness of the method, simulations were conducted using three heterogeneous trains for virtual coupling target tracking. The results indicate that compared to fixed-parameter MPC, the method proposed in this paper improves computational efficiency by 5% and reduces inter-train spacing tracking error by 16%. The effectiveness of the proposed control approach has been demonstrated, providing a reference for further research on dynamic adjustment of control parameters in train operation processes.