A convolutional neural network model for abnormality diagnosis in a nuclear power plant

Diagnosing abnormal events in nuclear power plants (NPPs) is a challenging issue given the hundreds of possible abnormal events that can occur and the thousands of plant parameters that require monitoring. This study proposes a convolutional neural network model for abnormality diagnosis in an NPP. The distinct feature of the proposed approach is the use of two-channel two-dimensional images to deal with (1) the massive amount of data that individual systems generate in real time, and (2) the dynamics of the states of individual systems. One channel represents the current NPP state values, while the other channel represents the changing patterns of the state values during a prescribed time period in the past. Experimental results from a full-scope simulator confirm, with statistically significant outcomes, that the developed model outperforms other classification models in terms of accuracy and reliability and is robust across different contexts of analysis, and thus has the potential to be adopted by actual NPP systems for real-time diagnosis.