Abstract: The variety of sound sources and the complex acoustic environment in UHV transformer substations bring great challenges to the state evaluation and fault detection of UHV shunt reactors based on acoustic signals. In order to solve this problem, this paper proposed a noise extraction and separation technique of target reactor based on the microphone array, which is used to extract the acoustic signal of the target reactor body. Firstly, we separated the non-homonymous noise of the target reactor through a mask-based beamforming algorithm, then the co-directional noise of the target reactor was separated by the underdetermined blind source separation algorithm so as to realize the effective separation of complex noise in the substation. Through a verification test in the shielded room, it is proved that the error rate of this method is less than 10%, which meets the needs of engineering applications. The proposed method solves the difficulty of on-site noise separation and extraction in substation, which is of positive significance for the fault diagnosis of UHV shunt reactor.