Nuclear power plant sensor signal reconstruction based on deep learning methods

Abstract

In nuclear power plants (NPPs), sensors provide real-time monitoring for nuclear power plant operators and assist operators in decision-making. In transient or accident conditions, sensors may be unable to transfer data due to various failures. This could lead to incorrect operation by the operator and cause serious consequences. The authors proposed a signal reconstruction method based on a one-dimensional convolutional neural network (1D CNN). The sensor data set was collected from NPP, which represented a power-down process. In this study, the authors built, trained and tested the CNN’s reconstruction performance by reconstructing different signals in NPP. An experiment on steam generator water level was conducted to validate the robustness of the proposed model when multiple signals were missing. The authors used the error and standard deviation to evaluate the experimental results. The results showed that the proposed model has a good performance on NPP’s sensors.