Abstract: The power science and technology graph should be flexible to deal with the influence of ontology modeling, storage data marking and power entity recognition. A flexible construction of knowledge graph of power science and technology is studied. Firstly, the ontology with the technical programs as the core is built, upward correlation with technical system, horizontal correlation with research network, and downward correlation with technical entities are built. Secondly, this paper proposes to transform the entity labeling problem into a text classification problem, use the theme classification and the maximum entropy semantic classification method to solve the problem of literature entity labeling. In view of the new scientific and technological documents, the model construction under the new technology system is completed through the necessary data markers, so as to realize the knowledge of the new documents and the stock documents into the graph. Third, the accuracy of the power science and technology entities is improved through BiGRU-CRF. The above methods are integrated and verified based on the self-developed intellectual property digital management and control platform.