The power grid data is scattered and the degree of intelligence of the data is not high, which lead to the difficulty of constructing the knowledge graph from scratch. Moreover, cross-domain migration is difficult because of the large number of long-tail relationships in the knowledge graph. To solve above problems, this paper proposes a knowledge transfer adaptation algorithm for the power industry. By coding neighbor entities in the knowledge graph, the algorithm obtains as much information in the knowledge graph as possible. Our method selects reference sets by entity clustering, which can improve the richness of entity relation semantics in reference set and the accuracy of matching. In the process of entity matching, the dynamic change of task relational semantics in different situations is considered. The attention mechanism is added to assign different weights to the entity pairs in the reference set. The method achieves ideal results on power data, which verifies the effectiveness of the technique.