Scholarlink

Regarding the problem of low prediction accuracy for lifetime of SiC MOSFET/Si IGBT hybrid modules under unsupervised few-shot condition, making it difficult to accurately assess their health status. To solve these problems, a data augmentation method based on Improved Genetic Algorithm (IGA) and Generative Adversarial Network (GAN) is proposed in this paper, applied to hybrid module lifetime prediction under these conditions. The crossover and mutation probabilities of the genetic algorithm are dynamically adjusted according to an improved fitness function for the hyperparameter optimization of the GAN, enabling data augmentation for power device aging data. The Long Short-Term Memory (LSTM) networks are used for lifetime prediction. And an aging dataset of the collector-emitter peak voltage ( \({V}_{ce-p}\) ) of Si IGBT is used as an example to validate the effectiveness of the proposed method in this paper. Compared to the traditional GA, the parameter optimization efficiency of the IGA is increased by 33%, the data generation quality of the GAN is improved by 35.9%, and when used for lifespan prediction, the accuracy is enhanced by 19.79%.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Data Augmentation Method for Lifetime Prediction of SiC MOSFET/Si IGBT Hybrid Modules Under Few-Shot Condition

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Weixin Chen,
  • Zishun Peng,
  • Yuxing Dai,
  • Wen Hu,
  • Yongli Yi,
  • Shixi Lin,
  • Yao Zhou,
  • Huasen Xie

摘要

Regarding the problem of low prediction accuracy for lifetime of SiC MOSFET/Si IGBT hybrid modules under unsupervised few-shot condition, making it difficult to accurately assess their health status. To solve these problems, a data augmentation method based on Improved Genetic Algorithm (IGA) and Generative Adversarial Network (GAN) is proposed in this paper, applied to hybrid module lifetime prediction under these conditions. The crossover and mutation probabilities of the genetic algorithm are dynamically adjusted according to an improved fitness function for the hyperparameter optimization of the GAN, enabling data augmentation for power device aging data. The Long Short-Term Memory (LSTM) networks are used for lifetime prediction. And an aging dataset of the collector-emitter peak voltage ( \({V}_{ce-p}\) ) of Si IGBT is used as an example to validate the effectiveness of the proposed method in this paper. Compared to the traditional GA, the parameter optimization efficiency of the IGA is increased by 33%, the data generation quality of the GAN is improved by 35.9%, and when used for lifespan prediction, the accuracy is enhanced by 19.79%.