Power MOSFETs are critical components in power electronic systems. Predicting its remaining useful life is crucial to prevent catastrophic failures. In this paper, we propose a method for predicting the remaining useful life (RUL) of power MOSFETs based on U-Net time series prediction model (UnetTSF). First, noise in the measurement data was eliminated through mode decomposition and data reconstruction, and then the degradation trajectory and RUL of the power MOSFETs are predicted using the UnetTSF. Experimental verification is conducted using NASA's public dataset, with Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) selected as comparison algorithms. The experimental results indicate that the UnetTSF model achieves higher prediction accuracy and generalization, and performs well even with a small amount of data.