The invention provides an artificial intelligence identification method of external faults of a conversion current bus of an HVDC system. The artificial intelligence identification method is characterized in that a wavelet-transform multi-scale decomposition algorithm is utilized to decompose system fault signals into different frequency sections, different wavelet energy skewnesses of different frequency-section signals are used as fault diagnosis feature vectors, and the fault diagnosis feature vectors are combined with a BP neural network to achieve accurate identification of fault types of an AC system. Compared with an AC system fault type identification method in the prior art, the method only needs to extract characteristics of a fault signal from the aspect of energy distribution, is not affected by the system operation mode, and is simple in operation, good in accuracy, and high in reliability.