A multi-radial basis neural network-based partial discharge positioning error compensation method disclosed by the present invention comprises the steps of (1) dividing a space region; (2) acquiring asample partial discharge signal, and calculating a sample time delay value; (3) calculating a theoretical time delay value; (4) training a first radial basis neural network; (5) compensating an errorof the sample time delay value, and outputting a corresponding correction sample time delay value; (6) calculating a corresponding correction sample discharge source position; (7) training a second radial basis neural network; (8) acquiring a standby partial discharge signal at the discharge source position, and calculating a corresponding initial time delay value and an initial discharge sourceposition; (9) compensating an error of the initial time delay value, and outputting a corresponding correction time delay value; (10) calculating a corresponding initial correction discharge source position; (11) compensation an error at the initial correction discharge source position, and outputting a corresponding final correction discharge source position. The method of the present invention can compensate the time delay calculation errors and positioning errors, and enables the positioning precision to be improved.