The direction of arrival（DOA） estimation of multiple ultra-high frequency（UHF） partial discharge（PD）sources are difficult to implement. Based on the compressive sensing（CS） theory

with the steering vector matrix as the transformation matrix

with the random Gaussian matrix as the measurement matrix

the traditional 2×2 UHF array was virtually extended to a N×N（N>2） array using the regularization-focal underdetermined system solver to multiple measurement vectors（RM-FOCUSS） signal reconstruction algorithm. On the basis of wide-band signal focusing and environment noise suppression

the spatial spectrum of the PD signal sources was constructed using the multi-signal classification（MUSIC） algorithm

and the accurate orientation of the multi-local discharge source was realized according to the spatial spectral feature distribution. The simulation results show that the orientation accuracy of the extended 4×4 array is greatly improved and the directional angel error is reduced from about 10 degrees to 4 degrees. With the extended4×4 array in field tests as an example

the directional angle error is about 8 degrees and the pitch angle error is about 4 degrees. The number of partial discharge estimation has expanded from 1 to 3 compared to the traditional array

which further verifies the effectiveness of the proposed method. The research can provide a reference for the partial discharge detection of electrical equipment.