Abstract: The safe operation of distribution network is faced with challenges such as complex fault types, low positioning accuracy and insufficient efficiency of traditional methods. Based on this, an intelligent detection method combining multi-source signal acquisition, deep feature learning and fusion positioning mechanism is proposed. Experiments show that the fault type recognition rate of this method breaks through 98.7%, and the positioning error is less than 1%, which provides an effective new path for accurate fault diagnosis of smart grid.