Abstract: A rolling bearing is a key component of a rotating machinery,and the accurate prediction of its remaining useful life (RUL) can help maintenance personnel make maintenance plans in time,prolong equipment working time and ensure safety.Because it involves complex physical process to accurately establish a model of bearing degradation process through mathematical modeling,the data-driven method based on deep learning becomes a popular alternative method.This paper proposes an improved temporal convolutional network with hybrid dilated convolution and self-adaptive slope thresholding (TCN-HS) function to predict the RUL of rolling bearings.This model uses hybrid dilated convolution (HDC) to solve the problem of grid effect,and uses self-adaptive slope thresholding functions to further screen features.In order to verify the effectiveness of TCN-HS model,experiments are carried out based on PHM2012 bearing data set.The results show that the improved method upgrades the model and the prediction results are accurate.