<p>Predicting resource usage is crucial for managing cloud systems, ensuring that cloud services remain efficient and stable. However, the complex temporal patterns of resource sequences impose numerous limitations on prediction models in capturing their multi-dimensional features. Moreover, there are correlations between different types of resources in the cloud environment, and the dynamic patterns of resources are influenced by various external factors. Predicting future trends based solely on historical usage patterns is inaccurate. To this end, we propose WAPformer, a novel cloud resource prediction model that leverages wavelet decomposition in conjunction with attention mechanisms. The model employs wavelet networks to decompose resource sequences into multiple components, decoupling multi-dimensional potential influencing factors. It then integrates local and global attention mechanisms to capture feature patterns induced by potential factors. Finally, it combines self-attention mechanisms with external shared memory units to accurately model the dependencies between different types of resources. Experimental results demonstrate that WAPformer outperforms existing methods in cloud resource usage prediction.</p>