Traditional fault detection methods for power generation systems use cen-tralized fault processing analysis, which leads to long accuracy and responsetime of fault detection. To address these problems, a data mining-baseddistributed power generation system fault artificial intelligence detectionmethod is studied. The depth-first search tree algorithm is used to divide thegrid of distributed generation system. The network structure is modified tolocate fault zones by processing anomaly mining of the system data aftergrid division. The combination of fuzzy logic and wavelet singular entropyis used to complete the detection and identification of system faults. Throughsimulation experiments, it is verified that the response time of the detectionmethod is only 0.016 s, and its detection error rate and false negative rate are1.23% and 1.25%, which are far lower than other methods.