<p>Entity alignment (EA) aims to identify and link equivalent entities between different knowledge graphs, which is crucial for integrating multi-source data and improving its usability. At present, mainstream EA methods focus primarily on traditional knowledge graphs, employing techniques such as embedding models, graph neural networks (GNNs), and semantic similarity. However, these approaches usually ignore the temporal information in the knowledge graphs, which limits the accuracy of the alignment. In addition, although relation representation has been shown to enhance entity representation and improve alignment, most existing methods rely on entity representation to learn relation representation, without fully considering the diverse topological structures among relations, such as adjacency and ring structures, which limits the enhancement of entity representation. To address these issues, we propose a novel entity alignment approach that integrates temporal information and relational structures. Specifically, we employ a time-aware graph attention network (TGAT) to aggregate the temporal relations of adjacent nodes and assign weights to nodes based on the relevant temporal information. In parallel, we use Line Graph (LG) modeling to capture relational topologies such as adjacency, rings, and reciprocity. This enables relation representations to be learned independently from entities, thereby enriching entity embeddings. Finally, we formulate the alignment problem as an assignment problem and solve it using a sparse Sinkhorn iteration. Experimental results show that our approach significantly outperforms existing methods on multiple real-world Temporal Knowledge Graph (TKG) datasets.</p>