Considerations on Data Mapping of Convolutional Neural Networks to Diagnose Abnormal Status in Nuclear Power Plant Operation

Operator should be trained to diagnose the case of abnormal status to maintain the plant safe. However, due to various abnormal cases, operational mental workload to keep knowledge for diagnosis is relatively high. To reduce the mental workload, operation support system based on artificial intelligence needs to be developed. There are many algorithm in artificial intelligence model. The CNN is one of promising algorithm for diagnosis of abnormal status. The data mapping methodology to represent abnormal status efficiently is also need to be defined. The sensitivity test is planed according to the proposed considerations. Followings are major factors for the accuracy. 1) Degree of abnormality, 2) Type of abnormal status (affected variable range), 3) CNN parameters, 4) Characteristic mapping configuration. Proposed methodology with the CNN is bases on snapshot of plant status. However the changes of plant status is also important characteristic of abnormal plant status. When the sensitivity test has been completed, the necessity of application of status changes should be reviewed.