Coherent cluster identification plays an important role in the dynamic equivalent and active splitting control of power systems. This paper presents a coherency clustering method based on Prony analysis feature extraction. First, given the disadvantage of Prony analysis which is seriously disturbed by noise, the method of Ensemble Empirical Mode Decomposition (EEMD) is used to reduce the noise of the signal. Then, Prony analysis is performed on the angle signal to extract feature values such as amplitude, frequency, and damping to form the feature vector of each generator. Then the feature matrix formed by all the generator feature vectors in the system are input to a self-organizing neural network to realize coherency clustering. Taking the EPRI-36 bus system and North China grid as case studies, it is verified that this method can effectively eliminate the impact of noise on Prony analysis, fully extract the feature of the power angle signal, and accurately identify the coherency generator group. This work is supported by National Natural Science Foundation of China (No. 51777067) “Research on Dynamic State Estimation of Comprehensive Energy System Based on Bond Graph Theory” and Independently Introduced the Scientific and Technological Project of State Grid Economic and Technology Research Institute Co., Ltd. (No. 524414190002) “Dynamic Performance Research of HVDC Transmission System Based on Electromechanical Transient-Electromagnetic Transient Hybrid Simulation Technology”.