Degradation prediction method of switching power supply based on EMD-LSTM

As a crucial energy provider in electronic systems, accurate monitoring and assessment of the health status of switching power supplies are crucial for ensuring efficient system operation. This research suggests a hybrid strategy that combines empirical mode decomposition (EMD) with a long short-term memory (LSTM) network to overcome the shortcomings of both data-driven and classic physics model-based methods in predicting complicated, non-stationary degradation signals. In the proposed method, EMD is employed to decompose degradation signals into multi-scale intrinsic mode functions. The relevant components are then selected and further processed, while the LSTM network captures long-term dependencies and nonlinear temporal dynamics in the time series. This approach enables accurate prediction of degradation trends in switch-mode power supplies. To verify the effectiveness of the proposed algorithm, a degradation simulation test platform for a switch-mode power supply is designed, and the fault injection method is developed. In order to simulate component-level degradation processes realistically, a fault-injection circuit is expressly made to mimic the degradation behavior of important components, especially capacitor degradation within the filtering module of the switch-mode power supply. Based on the constructed degradation experimental setup, experiments are conducted under degradation conditions to systematically validate the effectiveness of the proposed method.