【目的】针对大规模直驱风电场详细电磁暂态建模面临“维数灾”导致计算负担过重，以及传统单机等值模型难以精确反映次同步振荡特性的问题，提出了一种高精度、高效率的风电场动态等值建模方法，以有效支撑风电并网系统次同步振荡特性分析，并揭示整个风电场振荡特性与关键振源机群特性之间的内在关联。【方法】基于风电并网系统的特征值分析计算主导特征值参与因子，辨识影响次同步振荡特性的关键因素，并基于K-means聚类算法，提出将运行工况、集电线路与系统状态变量稳态初值三者作为聚类指标的风电场分群动态等值建模方法，推导并给出了相应的分群等值参数计算方法。【结果】搭建30台直驱风机联网系统的电磁暂态仿真算例，从特征值、时域仿真结果对比，多角度验证了用于次同步振荡分析的分群聚合等值模型的有效性。【结论】所提出的基于主导特征值参与因子指导、融合运行工况、集电线路及状态变量初值三指标聚类的风电场多机分群等值建模方法，可显著提高次同步振荡分析的精度和效率，突破了单机等值局限，能够精准捕捉并分离风电场内部不同集群（特别是振源机群）对次同步振荡的贡献，可为大型直驱风电场并网系统的次同步振荡机理研究、稳定性评估及抑制策略制定提供可靠的建模工具。

[Objective] A high-precision and high-efficiency dynamic equivalent modeling method for wind farms is proposed to address the heavy computational burden caused by the curse of dimensionality in detailed electromagnetic transient modeling of large-scale direct-drive wind farms and to overcome the limitations of traditional single-machine equivalent models in accurately reflecting sub-synchronous oscillation characteristics. The method enables effective analysis of sub-synchronous oscillations in grid-connected wind power systems and reveals the inherent relationship between the oscillation behavior of the entire wind farm and that of key vibration source groups. [Methods] Based on an eigenvalue analysis of wind power grid-connected systems， the dominant eigenvalue participation factors were calculated to identify the key factors affecting the sub-synchronous oscillation characteristics. Based on the K-Means clustering algorithm， a dynamic equivalent modeling method for wind farm clustering is proposed， which uses the operating conditions， collection lines， and steady-state initial values of the system-state variables as clustering indicators. The corresponding clustering equivalent parameter calculation method was derived and presented. [Results] An electromagnetic transient simulation example was built for a networked system of 30 direct-drive wind turbines. The effectiveness of the clustering and aggregation equivalent model used for the sub-synchronous oscillation （SSO） analysis was verified from multiple perspectives by comparing the characteristic values and time-domain simulation results. [Conclusions] The proposed multimachine clustering equivalent modeling method for wind farms based on dominant eigenvalue participation factor guidance， fusion of operating conditions， collection lines， and initial values of state variables can significantly improve the accuracy and efficiency of SSO analysis， overcome the limitations of single-machine equivalence， and accurately capture and separate the contributions of different clusters （particularly vibration source clusters） within the wind farm to sub synchronous oscillations. This can provide a reliable modeling tool for the study of sub-synchronous oscillation mechanisms， stability evaluation， and suppression strategy formulation for large-scale direct-drive wind farm grid-connected systems.