Fault detection for nuclear power plant based on improved moving window and sparse autoencoder

Due to factors such as component performance degradation and changes in operating conditions, nuclear power plants (NPPs) equipment exhibits significant time-varying characteristics during operation, leading to the failure of fault detection models. Therefore, this study proposes a fault detection method based on an improved moving window and sparse autoencoder to enhance the adaptability of the detection method to the time-varying data of NPPs. This method establishes a sparse autoencoder as a fault detection model, determining the operating status of equipment by calculating the statistical relationship between test data and reconstructed data. In this process, the traditional moving window update strategy is optimized based on Euclidean distance, and the improved moving window strategy enables effective model updating. Finally, the effectiveness of the proposed method is verified using data from a nuclear power plant reactor coolant pump. The results show that the proposed method performs well in terms of fault detection rate and false alarm rate.