Temporal Knowledge Graph Completion (TKGC) aims to address the incompleteness issue present in Temporal Knowledge Graph (TKG). Existing methods for TKGC mainly fall into two categories: one is the method that combines temporal information with entity and relation representations, which makes it difficult to deal with complex temporal patterns, and the other is the method that uses Graph Neural Network (GNN) to capture the neighborhood structure, which usually focuses on single timestamps and ignores the interactions between different timestamps. To address these limitations, we propose a novel method called Dynamic Periodicity Perception and Multi-Graph Integration (DPPMI). DPPMI introduces Temporal Category Sampling strategy and Relation-Aware Graph Transformer module to effectively capture contextual information across different time points. To handle complex temporal dynamics, we introduce a novel period embedding method based on the prime. Furthermore, we introduce a specialized attention mechanism to dynamically perceive the significance of various period embeddings, enabling the model to effectively identify and capture complex temporal patterns. Experimental results show that our model improves the Mean Reciprocal Rank (MRR) on the ICEWS14, YAGO11k, and Wikidata12k datasets by around 3.9%, 10%, and 3.7%, respectively, compared to the state-of-the-art baseline.