Remaining Useful Life Prediction Based on an Improved Residual and Bi-directional GRU Network

Aiming at the problems of difficult to establish feature extraction model and low accuracy of remaining useful life (RUL) prediction, a RUL prediction model based on improved residual and bi-directional gated recurrent uint (Bi-GRU) was proposed. Firstly, the key data of the motor data set is preprocessed to provide the data basis for the RUL prediction model. Secondly, the improved residual neural network is used to fusion and feature extraction of multi-sensor signals to obtain the health index (HI) of the device. Finally, the obtained HI is input into Bi-GRU to obtain the predicted value of the RUL, and the effectiveness of the proposed method is verified by the open data set.