With the continuous improvement of the level of power grid informatization, the traditional power system contains a considerable amount of unstructured raw data, and current models cannot handle the unstructured text data well, which hinders the digital transformation of the power system and the improvement of work efficiency. Firstly, the propagation mechanism of different fault types of the main network line is analyzed, and the fault analysis and research of multi-physics combined numerical simulation are proposed in time to extract the key electrical features of typical fault-type scenarios. Secondly, a deep learning model built on the CNN-BiGRU-CRF is proposed to extract knowledge from multi-source heterogeneous power data, effectively addressing the issue of poor entity recognition effect of small-sample data in the power field. Thirdly, an advanced artificial intelligence-based decision support system for managing main network faults is introduced, and the knowledge graph of main network fault scheduling and disposal is established so as to transform the data in the multi-source heterogeneous power field into knowledge. Finally, the design experiments prove the effectiveness of the auxiliary decision support system function of the main network line fault disposal proposed in this paper. The model’s accuracy reaches 93.26%, and the F1 value reaches 95.58%. Finally, the knowledge graph established by the Neo4j graph database was used to display the knowledge graph visually, and its application in the auxiliary decision support system of mainnet fault handling and scheduling was analyzed.