Research on Degradation Prediction Technology of Secondary Switching Power Supply in High-Temperature Aging Test System Based on Liquid Neural Network Fusion Architecture

Aging test bench can accelerate the aging process of electronic components, due to the single aging time up to 1000 hours and above, in order to enhance the reliability of the aging test results. In this paper, the key component of the aging test bench switching power supply, that is, the secondary power supply module as a research object. The data sampling rate generated by the secondary power supply is extremely high, and the data scale expansion brought about by the high sampling rate data, together with the limited computational resources of the model brings great challenges to the degradation analysis and prediction of the secondary power supply. For this reason, this paper proposes an innovative prediction method that can reduce the amount of computation while maintaining the prediction accuracy. First, a sliding window is used to process the high sampling rate data, based on which a model fusion architecture based on liquid neural network (LNN) is constructed, which deeply fuses the long shortterm memory network (LSTM) and temporal convolutional network (TCN).The LSTM is good at capturing the long-term dependencies in the time-series data, whereas the TCN can efficiently process time-series data by virtue of its causal and expansive convolutional mechanisms. processing time series data with its causal convolution and dilation convolution mechanisms. Liquid neural networks, as the core of model fusion, enhance the fusion model's ability to capture features at different time scales and adapt to complex dynamic changes. Finally, the prediction of voltage profiles by model fusion provides a more accurate basis for the degradation prediction of aging stations. Experimental results using NASA public datasets combined with physical simulation models show that the method has obvious advantages over traditional methods.