Attention-Based Variational Encoding Regression Method of Predicting Engine Remaining Useful Life

Accurate remaining useful life (RUL) prediction result for turbofan engine is critical to aircraft safety, however, the complex operation conditions and high-dimensional monitoring data seriously affect the prediction performance. To address the problems, this article presents an attention-based variational encoding regression model. The proposed method adopted variational auto-encoder whose total loss function jointly considers the reconstruction error and the prediction error for improving the feature exaction ability and prediction performance. furthermore, bi-directional gated recurrent unit (Bi-GRU) neural network and attention mechanism are adopted to exact and focus on the important hidden features effectively. Experiments based on four datasets are conducted to assess the effectiveness of the model, compared to other methods, this method can achieve superior performance.