Development of a generative-adversarial-network-based signal reconstruction method for nuclear power plants

Abstract

In nuclear power plants (NPPs), the reliability of instrumentation signals is crucial for making appropriate decisions. Multiple signals can become faulty simultaneously and become unavailable under harsh conditions, and this could lead to improper decisions being made, which could result in catastrophic failure. The authors proposed a new signal reconstruction method based on a generative adversarial network (GAN) that can be applied to reconstruct multiple missing signals. In the method, a GAN model is first trained to generate realistic signal sets from random latent vectors and then used to find the optimal latent vector for the reconstruction of a damaged signal set. The signal set is reconstructed by replacing its missing parts with the corresponding parts of the signal set generated from the optimal latent vector. Experiments were conducted to verify the applicability of the proposed method to the reconstruction of multiple missing signals under various NPP emergency situations.