Aiming at the difficulty in establishing an accurate physical model of a power converter that represents the degradation process, a method based on unscented particle filter (UPF) is proposed in this paper to realize the remaining useful life (RUL) prediction. First, through the analysis of influences on the circuit performance due to the degradation of key circuit components, the average output voltage is selected as the characteristic parameter of useful life. Then, UPF is used to perform modeling on the fault trend based on the history data of circuit performance degradation. Finally, the RUL prediction of the power converter is realized by step-by-step recursion of characteristics with the combination of the circuit’s failure threshold. A closed-loop SEPIC circuit is taken as an example, and the influences of modeling data size on the prediction performance are analyzed. In addition, the effectiveness and accuracy of the proposed method are verified in comparison with the Kalman filter (KF) method.