XLSTM-Informer-Based Task Load Prediction in Cloud-Edge-End Collaborative Architectures

In cloud-edge-end collaborative computing architectures, traditional models often fail to accurately predict the dynamic, heterogeneous, and nonlinear task loads, which negatively impacts resource scheduling efficiency and increases system energy consumption. To address these challenges, this paper proposes a hybrid deep learning model, Extended long short-term memory (XLSTM)-Informer, which combines the advantages of XLSTM and Informer. This model captures both local dependencies in long sequence data and global temporal features. Specifically, XLSTM incorporates scalar Long Short-Term Memory (LSTM) and matrix LSTM to enhance long-sequence modeling through exponential gating, matrix storage, and a key-value retrieval mechanism. Informer employs probabilistic sparse attention and a distillation mechanism to significantly reduce computational complexity. Experimental results using the Alibaba Cluster Trace-v2018 dataset demonstrate that XLSTM-Informer outperforms existing models, achieving a 37.4%, 85%, and 86.9% reduction in mean absolute error compared to Informer, LSTM, and Convolutional Neural Network (CNN), respectively. These results verify the superior accuracy and robustness of the model in complex dynamic environments.