Abstract: In order to solve the problems of lagging rule updates and limited fault identification capabilities caused by the single data source of the traditional operation and maintenance knowledge base and the need for the experience of operation and maintenance personnel, an intelligent operation and maintenance knowledge base system based on deep learning is constructed.The system integrates the fault classification model and the twin network text detection model, and combines the knowledge graph to construct a four-layer architecture of data input layer, data storage layer, knowledge inference layer, and application layer, realizing automatic mining of knowledge from multi-source data and dynamic update of auxiliary rules.The fault classification model adopts the deep convolutional neural network (DCNN) algorithm, and the rule update adopts the twin network text detection model, which is verified by combining experiments.The results show that the proposed system has an accuracy rate of 99.6% in the recognition of operation and maintenance images and 92.6% in short text matching, which effectively improves the efficiency of equipment operation and maintenance and the level of knowledge management.