In order to solve the problem of poor accuracy and low efficiency of traditional rolling bearing life prediction method. A prediction method based on bi-directional stack simple recurrent unit is proposed. Multiple time and frequency features are extracted from the original signal to construct a multi-dimensional data set to enhance the information representation and avoid the insufficient reflection of a single feature on the bearing degradation ability. Relying on the ability of recurrent neural network to process time series data,the SRU model is constructed,the traditional RNN serial operation structure is reconstructed,and the training efficiency is improved. On the basis of single layer SRU,it is stacked in reverse,so that the model can learn two-way information in time dimension,improve the accuracy of prediction. The experimental results show that the prediction accuracy and training efficiency of this method are improved.