Research on Mechanical Equipment Remaining Useful Life Prediction Method Based on Attention Mechanism and Feature Fusion

This paper proposes a deep learning method combining multi-head attention mechanisms and multi-source feature fusion for remaining useful life (RUL) prediction of mechanical equipment. The approach extracts temporal dynamic features through a Long Short-Term Memory (LSTM) network and introduces a multi-head cross-attention module to dynamically allocate weights to temporal features. Additionally, a feature enhancement module is designed to integrate handcrafted features with CNN-extracted local features, enhancing the model’s capability to characterize equipment degradation patterns. Experiments on the C-MAPSS FD004 dataset demonstrate that the proposed method achieves RMSE and Score metrics of 22.64 and 4723.19, outperforming LSTM-Attention methods by 19.3% and 39.8%. The research provides a novel solution for RUL prediction under complex operating conditions.