Abstract: According to the requirement of array signals de-noising,it needs to reserve the information of "phase difference" in the maximum limit.The author utilized complex FastICA algorithm to de-noise the partial discharge（PD） ultrasonic array signals containing noise in this paper.Firstly,the author realized the successive separation of source signal and noise signal in mixed signals through the complex separable matrix U;then adopted the waveform similarity principle to extract separated signals blindly,and got the PD ultrasonic signals after the de-noising;on this basis,according to the sensor array model,the author made physical reconstruction and got array signal model.At the same time,taking array flow pattern’s average phase angle difference（PAD） as a core,combining the variational trend parameter（VTP）,normalized correlation coefficient（NCC） and the signal to noise ratio（SNR） before and after the de-noising to evaluate the simulation PD ultrasonic array signals synthetically.Finally,the author utilized the FastICA algorithm to de-noise the actual measured PD ultrasonic array signals in the lab.The results show that FastICA algorithm can restrain the white noise in PD ultrasonic array signals,with the premise of high SNR,it can reserve the information belonging to phase difference of array signals in the maximum limit and ensure the good result of direction and location.