Abstract: The fault handling pre-plan of power grid has great instructive significance for the quick emergency disposal when the failures or accidents occur. To assist dispatchers in fault handling and improve the capabilities of power grid's emergency handling and the level of dispatch intelligence, the technology of knowledge graph can be used to extract, represent, and manage fault handling pre-plan. By exploring the fault handling pre-plan of the power grid, this paper proposes a new method to construct the knowledge graph for the fault handling.. The method combines both the top-down and bottom-up construction strategies and solves the involved problem of the knowledge extraction for the power domain. Firstly, the scheme layer of the knowledge graph in the top-down style is designed, which defines the knowledge framework, the concept types, and the relationships between the concepts of the knowledge graph. Then, according to the characteristics of the pre-plan text, multiple deep learning models are comprehensively used for knowledge extraction, and build the data layer of the knowledge graph in the bottom-up style. To avoid word segmentation errors, the TextCNN model is used based on character-level vectors to classify the text content of the plan. For the word conflict, the LR-CNN model is applied to identify domain named entities. On the basis of named entity recognition, the BiGRU-Attention model is adopted to extract the relationships between the entities. Finally, the effectiveness of the above-mentioned knowledge extraction method is verified through experiments. The constructed power grid fault handling knowledge graph is visualized and its application in intelligent information retrieval and auxiliary fault diagnosis is analyzed.