Model predictive control (MPC) is applied in modular multilevel converter (MMC) control due to its advantages of fast response and simple modeling. However, previous research has not addressed high-frequency resonance suppression. Firstly, a mathematical model for MMC predictive control is established. By combining the dq impedance modeling method in traditional control, the Z-transform is used to achieve positive and negative sequence impedance modeling of the MPC. Secondly, the discrete transfer function of the converter station body and conduct stability analysis are derived. Under the premise of a stable ontology, the impedance method is used to reveal the mechanism of the station high-frequency resonance in model predictive control. Then, in response to the high-frequency resonance problem caused by delay, a multistep predictive control method is used to suppress it. Based on the impedance model, the optimization effect on the local high-frequency region is analyzed. Finally, the theoretical analysis shows that the converter station can compensate for the increased delay by increasing the number of predicted steps. And the effectiveness and correctness of multi-step predictive control in suppressing high-frequency resonance are proved through electromagnetic transient simulations.