Application of deep learning techniques for nuclear power plant transient identification.

• Timely and correct identification of NPP transients could either prevent an accident or mitigate the consequences of an accident. • A hybrid deep learning technique is proposed for the online identification of nuclear power plant transients. • A CNN-LSTM network was used for the transient identification, which transformed the input matrices into vectors using a CNN and then LSTM layers predicted the probability of each transient by sequence processing. • The results were shown that the proposed technique performs better than common deep learning techniques in terms of accuracy, identification time, and computational cost. Identification of NPP transients plays an important role in the prevention of accidents and mitigation of their consequences. NPP parameters may follow different patterns during each transient. So the transients can be identified by monitoring the operating parameters. It has been shown in several studies that data-driven methods, especially deep learning approaches, have a desirable performance in NPP transient identification. A hybrid deep learning technique is proposed in the present paper, in which transient identification is done using a CNN-LSTM neural network. The training data set is taken from a VVER-1000 full-scope simulator and the most important operating parameters are determined by feature selection techniques. According to the results, the proposed technique has identified the NPP transients in a short time, with high accuracy, and with a reasonable computational cost. The effective performance of the technique makes it possible to use it as a practical tool for online transient identification. [ABSTRACT FROM AUTHOR]