Temporal knowledge graph (TKG) logical reasoning aims to predict future facts by learning logical correlations from TKGs. However, conventional Retrieval-Augmented Generation (RAG) methods rely on unstructured text as the retrieval space, failing to capture and utilize implicit logical correlations within the retrieved content. Furthermore, the conventional rule confidence calculation methods significantly restrict the generalization of the model when temporal logical rules are incorporated. To address these challenges, We propose a novel TKG reasoning framework RuleRAG (integration Rule with RAG). It validates the confidence of temporal logical rules using large language models (LLMs), and applies the rules and the query as input to LLMs. Experimental results demonstrate that RuleRAG outperforms state-of-the-art models.