Abstract: The defect handling work of power equipment mainly depends on the knowledge reserve and experience of the handling personnel. However, due to the lack of assistance of a perfect historical defect knowledge base, people with relatively insufficient experience and knowledge cannot effectively learn from the experience of the predecessors, and it is inevitable that there will be mistakes in decision-making. This situation, which in turn affects the elimination of power equipment. A question answering system implementation method based on power equipment defect knowledge graph is proposed to address the above issues. Firstly, the requirements of the equipment defect question answering system are analyzed, and the system architecture is designed. Then, the question entity recognition model, question intent recognition model, and query sentence generation are established respectively. The model parses the question sentence semantically. Finally, a power equipment defect question answering system is based on the power equipment defect knowledge graph. The results of question entity recognition and question intention recognition show that the indicators of the improved algorithm have been greatly improved. In the aspect of question entity recognition, the precision rate, recall rate and F₁ reach 92.34%, 97.65% and 95.36%, respectively. In the aspect of question intention recognition, the accuracy rate, precision rate, recall rate and F₁ reach 82.17%, 85.38%, 82.36% and 80.56%, respectively. The function test of the question answering system also shows that the system can be well applied to the defect elimination process of power equipment. And the system can quickly improve the accuracy of defect, repair strategy formulation and the efficiency of defect elimination of defective equipment, and promote the safe and stable operation of the power system.

Key words: question answering system, power equipment, knowledge graph, question entity recognition, question intent recognition, query statement generation