Dynamic Energy Sparse Self-Attention Based on Informer for Remaining Useful Life of Rolling Bearings

Predicting the remaining useful life of rolling bearings presents significant challenges due to their nonstationary degradation patterns. Conventional methods struggle with long sequences and are strongly dependent on manual feature extraction, which can lead to inefficiencies and biased predictions. To address these issues, this paper proposes the dynamic energy sparse self-attention model based on Informer. The proposed model uses the auto-reassignment transform to extract more accurate features from the raw vibration signals, and the dynamic energy sparse self-attention mechanism adapts the attention weights based on the energy distribution. In addition, the Monte Carlo dropout method is employed to quantify uncertainty and enhance model robustness. Experimental results obtained on two benchmark datasets demonstrate that the proposed model outperforms existing advanced methods, achieving a reduction in mean absolute error and root mean square error, demonstrating its superior ability to handle complex bearing degradation prediction tasks.