Abstract: Aimed at the problem of low localization accuracy and long delay of transformer partial discharge sound plot position, a transformer partial discharge sound plot positioning method based on time-domain audio separation network (TasNet) and neural generalized cross-correlations (NGCC) is proposed in this paper. Firstly, the features of the audio sequences are identified and separated through the mic-array and the TasNet network; then, the estimated time differences of arrival (TDOA) value corresponding to the partial discharge sound plot based on the convolutional neural network are obtained; finally, a positioning framework to calculate the location information of the partial discharge sound plot is built to achieve localization of transformer partial discharge sound plot. Experiments show that compared with the traditional method based on generalized cross-correlation function with phase transform (GCC-PHAT), the method proposed in this paper significantly improves the accuracy and efficiency of partial discharge sound plot position.