The present invention proposes a fault discrimination and reasoning method based on knowledge graph, which acquires equipment data, fault data, and disposal plan data; establishes an equipment map based on equipment data, establishes a fault map based on fault data, and establishes a disposal plan map based on event data; The extraction algorithm and the TranSE algorithm perform map fusion, map completion, and map reasoning on the equipment map, fault map, and disposal plan map to obtain a knowledge map; use the graph neural network to perform fault discrimination and reasoning on the knowledge map; use a large amount of equipment data and historical faults The data and the treatment plan data are respectively established atlases, and then the atlases are fused and completed, and the final knowledge atlas is used to perform fault discrimination and reasoning, and automatically provide methods for discrimination, diagnosis, and maintenance for transformer faults. The method is accurate in diagnosing faults Good performance, timely and reliable.