ETSEnet: Efficient Transformer with Squeeze and Excitation for Turbofan Engine Life Prediction

Predicting the Remaining Useful Life (RUL) is vital for the Prognostic Health Management (PHM) of turbofan engines. Recently, deep learning models like Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) networks have been applied to RUL prediction. However, these models often struggle to balance the ability to capture long-range dependencies with computational efficiency. To address this, we propose an efficient RUL prediction method named ETSEnet, based on an improved Reformer with multi-channel perception. This model utilizes a hashing bucket strategy for sparse computation, reducing computational resource consumption while capturing long-term dependencies. Additionally, the Squeeze-and-Excitation mechanism quantifies inter-channel correlation and enhances dependency through adaptive feature recalibration, significantly improving prediction accuracy and convergence speed. Experiments on the NASA C-MAPSS dataset demonstrate that our model surpasses others in both accuracy and training time.