In order to directly use neural network to automatically extract feature information from the collected full-life vibration signals and avoid the dependence on manually extracted features， a remaining useful life prediction method based on convolution gated recurrent unit （ConvGRU） attention is proposed. Firstly， the collected equipment vibration signal is input into ConvGRU-attention model after preprocessing. ConvGRU extracts the spatial local features of equipment state through convolutional neural networks（CNN） and gate recurrent unit （GRU） extracts the timing feature information， so that the equipment state features can be extracted more effectively. Secondly， the attention mechanism is used to assign different weights to the feature information. Then， the visualization experiment of the feature output of the intermediate network layer is carried out， which verifies the effectiveness of the feature extraction of this research method. Finally， experiments are carried out on two mechanical equipment datasets PHM2012 bearing dataset and NASA engine dataset， and compared with existing methods. The experimental results show that the prediction accuracy of the remaining useful life prediction method based on ConvGRU-attention is better and has better generalization.