Large amounts of data resources including relevant standards, products and technical documents, document management, fault recover records, etc. in the power domain continue to accumulate. How to fast query and search of these documents has important value for grid scheduling and fault recovery. The traditional search system is based on the key words matching, which cannot find accurate answers for query business terms. This paper designs a semantic search system for power domain. We research on word segmentation technology, knowledge graph and inference engine. The design architecture and key modules of the system are introduced, and the effectiveness of the method is evaluated by experiments.