In the context of smart grid, aiming at the condition monitoring of conventional circuit breakers with complex mechanical actions, an optimization method for remaining life prediction of conventional circuit breaker based on deep learning is proposed. Firstly, the variational mode decomposition (VMD) is used to decompose the opening vibration signal, and the mode with larger kurtosis is selected for reconstruction to highlight the effective shock characteristics of the signal. Then, the feature attention convolutional neural network (FACNN) is introduced for life prediction, and the feature attention module is embedded in the one-dimensional convolution layer to optimize the ability of neurons to capture key state information. Finally, the measured data of the circuit breaker is used for verification. The results show that the method can realize the prediction of the remaining mechanical life of circuit breakers in a targeted manner, and has a high prediction accuracy and stability, which effectively reduces the influence of data uncertainty caused by the complexity of the system.