Employing the Generative Adversarial Networks (GAN) for Reliability Assessment of Converters

Mission profiles are widely used for the reliability analysis of power converters. Typically, to assess the converter reliability, long-term (e.g., one year) mission profiles are adopted, and it is assumed that the profiles will be repeated in future years. However, due to mission profile uncertainties, the assumption can introduce considerable errors in the estimated reliability. In this paper, the errors introduced by the above assumption are studied in detail. Furthermore, to tackle this challenge, the paper proposes using the Generative Adversarial Networks (GAN) to generate unique mission profile scenarios that capture the temporal and probabilistic properties of the real profiles. In this regard, the effectiveness of using the GA-generated profiles to improve the accuracy of the estimated reliability is demonstrated.