To address the challenges of long-distance dependencies and environment diversity in cloud server load anomaly detection, an improved sparrow search algorithm (ISSA) based TCNFormer (ISSA-TCNFormer) is proposed in this paper. Initially, we take the advantages of TCN for local feature extraction and the global perceptron of Transformer, and propose the cloud server load anomaly detection network TCNFormer. Sparse self-attention mechanism is introduced into the vanilla Transformer to reduce the interference from redundant attentions in long-time series analysis. To further optimize parameter selection during the training process, an improved sparrow search algorithm is proposed to automatically finetune the training parameters. We conducted comprehensive comparative experiments and ablation analysis on the AIOpts Challenge dataset. The experimental results show that ISSA-TCNFormer achieved the best performance, proving the effectiveness of the proposed method in cloud server load anomaly detection.
|