In order to establish a more accurate wind farm equivalent model, a direct-drive wind farm clustering method based on principal component analysis is proposed. Firstly, all the state variables that characterize the operating state of the wind turbine are obtained by modeling and analyzing the direct-drive wind turbine. Then, three dominant variables are extracted by principal component analysis, which represent more than 90% of all state variables. It can accurately reflect the operating point of the wind turbine. Finally, for comparative analysis, the hierarchical clustering algorithm is used for clustering calculation with three dominant variables and wind speed as the grouping index. The detailed model of wind farm, the equivalent model of dominant variable grouping and the equivalent model of wind speed grouping are built on Matlab/Simulink platform. By comparing the simulation results of the dynamic characteristics of the three models under wind speed fluctuation and grid fault condition, the correctness and high accuracy of the proposed clustering method are verified. The wind farm equivalent model established by this method has higher accuracy. This work is supported by Natural Science Foundation of Hebei Province (No. E20185021134) and Science and Technology Project of State Grid Headquarters:Research on Inertia, Damping and Primary Adjustment Method of Doubly-fed Induction Wind Turbine (No. SGTYHT/18-JS-206).