Online Short-Term Aging Status Prediction of SiC MOSFETs for DC Solid-State Power Controller using Adaptive Variable Time-Steps Metabolic Gray Model

SiC MOSFETs are the key ideal device for the DC Solid-State Power Controller (DC-SSPC). The reliability of DC-SSPCs can be improved by online predicting of SiC MOSFETs aging degree. Among current prediction methods, numerous training data or precise physical parameters are required to establish prediction models, which will consume lots of computing resources or implement abundant aging experiments. GM(1,1)-based grey prediction methods have low requirements for data volume and system evolution law, making it more suitable for online prediction. However, SiC MOSFETs' aging data show the strong nonlinearity, leading to the low prediction accuracy of the traditional GM(1,1). In this paper, addressing authors' past work shortcoming in terms of the aging feature parameter with the high nonlinearity, the improved low-nonlinearity aging feature parameter is proposed. Furthermore, considering that the entire life cycle of SiC MOSFETs can be divided into different aging zones using the proposed aging feature, a segmented variable time-step metabolic GM(1,1) prediction method is proposed to further reduce the data series nonlinearity. Finally, the effectiveness of the proposed prediction method is verified based on the aging data of SiC MOSFETs for DC-SSPCs, compared to the traditional GM(1,1), its prediction accuracy is improved by more than 8%.