Current researches on partial discharge localization method which based on the time delay of ultra high frequency（UHF） signals mainly focus on improving the precision of signal time-delays

while the error correction of the localization system is seldom involved.We proposed a localization error correction algorithm based on multiple neural network. When the radius vector（r） ranges from 2 m to 6 m

6 m to 12 m

and 12 m to 18 m

respectively

a corresponding error compensation network was established. The radial basis function（RBF） neural network was trained by a finite number of time delay errors measured at the calibration points. The established neural network can simulate the error distribution and can be used to correct the actual localization results. The simulation and test results show that

by utilizing the error correction network

the localization accuracy can be improved and the discretization degree of the results can be reduced as well. The localization distance error can be controlled within 0.5 meters and the error of orientation angle is less than 6°

which can verify the error correction capability of the proposed algorithm.