In view of the problems of power switch cabinet such as complex operating environment, strong vibration disturbance, and needing to retain the vibration disturbance and analyze the characteristics of vibration disturbance, the Independent Component Analysis (ICA) de-noise method based on phase space reconstruction is applied to the partial discharge vibration signal of the switch cabinet of the power plant, and the signal separation is carried out. First, according to phase space reconstruction, this paper constructs the high dimension phase space matrix without destroying the dynamical characteristics of the signal itself. Then the independent component analysis method is used to separate the signal from the noise and extract the useful vibration signal. The application of the simulation signal and the actual data shows that the method is equivalent to or better than the wavelet method in de-noise. Besides, the optimal wavelet base and the optimal decomposition layer are not considered, and the self adaptation is strong and easy to implement. It is also suitable for de-noise of partial discharge vibration signals of switch cabinet of power plant. This work is supported by Key Science and Technology Project of Jiangsu Electric Power Company (No. J2017070).