Remaining useful life prediction of turbofan engine with GA optimized hybrid neural network

In recent years, deep learning has been widely used in industrial remaining useful life (RUL) prediction. However, most methods does not distinguish the importance of samples at different time steps. In order to solve this problem, this paper introduces an attention layer in neural network, and uses the multi-network collaborative training method to predict the remaining useful life. Convolutional Neural Networks (CNN) are used to extract the data information under the spatial span, and Bidirectional Gated Recurrent Units (BGRU) are used to capture the time correlation. Through parameter sharing and loss feedback, the improved genetic algorithm learns the attention weights of different time steps. The proposed method is validated on turbofan engine data and the experimental results show that the method outperforms several state-of-the-art methods.