Frequency response capability is necessary for wind farms (WFs) in the power system with a high proportion of renewable energy. In frequency control, WFs need to meet the grid’s power demand and avoid obvious additional damage to wind turbines (WTs) due to frequency control. In terms of meeting the grid power demand, the actual power characteristics of WTs need to be accurately described. An improved Gaussian Mixture Model (GMM) is used to model the actual measured output of WTs. The wind farm control center can accordingly issue achievable effective power commands to the WTs in the frequency response. In terms of avoiding the structural fatigue increase of WTs, a two-degree-of-freedom control strategy is designed, which simultaneously controls the generator torque and pitch angle and achieves fatigue suppression through multi-objective model predictive control (MOMPC). The simulation results show that the proposed control system can improve the frequency response capability of the WF and reduce the fatigue load of WTs, which has obvious application prospects.