Entity alignment, which aims to identify equivalent entities from various Knowledge Graphs (KGs), is a fundamental and critical task in knowledge graph fusion. Current entity alignment methods usually use relationship triples to represent the structural information of a KG and attribute triples to represent additional information beyond the structural information of a KG. However, the availability of such additional information is poor, and KG may not contain such it, so it is necessary to improve the entity alignment performance of KGs using only relation triples. In this paper, we propose a Multiscale Relational-gated Graph Network (MRGN) for knowledge graph entity alignment using only relational information. First, MRGN uses prompt learning to supplement the number of relation triples and assigns different weights to the relations in the supplemented relation triples and the original relation triples. Second, MRGN uses a multi-scale relational gating network to enhance the capture of long-tail entity information. Specifically, MRGN captures more entity neighborhood information and distinguishes between entities and relations. Experiments on four benchmark datasets show that MRGN outperforms state-of-the-art entity alignment methods.