Cross-document event coreference resolution aims to cluster events from multiple documents that refer to the same real-world events. Previous researchers have considered how to provide as much supplementary information as possible to the event representation from other aspects (such as event argument information) to help the model make correct judgments. However, they have only considered naive event mention features for the event itself, ignoring the key information features contained in the event itself in some cases. These important features need to be explicitly mined and utilized. In order to obtain a representation focused on the core features of the event, we first introduce word sense disambiguation to explicitly extract the information contained in the event. In order to enhance the event representation, we then use the Large Language Models (LLMs) to generate event explanations for comprehensive analysis and understanding. Additionally, we apply Rhetorical Structure Theory (RST) to parse the document and use GAT to derive the global event representation. Finally, the aforementioned information is fed into a multi-layer perceptron (MLP) to capture the similarities between event mention pairs for resolving coreferent events. The experimental results on both the WEC-Eng and FCC datasets demonstrate that our proposed method outperforms the state-of-the-art baselines.