Entity Alignment (EA) is the key to realize knowledge graph fusion. At present, the mainstream EA methods adopt the knowledge graph embedding technology under one and the same semantic space to obtain the entity and relation embeddings, and then the cost function is employed to measure the similarity of entity pairs. However, it is difficult to distinguish the different aspects of entities with different relations when representing a knowledge graph in the identical semantic space. Additionally, they ignores the multi-dimensional representation characteristics of topological structures, attribute information and semantic features in knowledge graphs. Therefore, this paper proposes an entity alignment framework that integrates the TransR model with multi-semantic features, and introduces the stable matching mechanism to overcome the defect of traditional methods that process entity pairs in isolation. Comparative experiments based on the DBP15K standard dataset show that the results of the algorithm is higher than the baseline models.