Using artificial intelligence to identify the success window of FLEX strategy under an extended station blackout

Abstract

The vulnerability of operational nuclear power plants (NPPs) to an extended station blackout (SBO) was revealed after the Fukushima Dai-ichi accident, which led to the development and implementation of new strategies such as the diverse and flexible (FLEX) strategy. The FLEX strategy has been adopted by several utilities around the world in support of existing emergency operating procedures (EOPs) to maintain the core coolability and hence ensure the plant safety under an extended SBO. In this paper an AI algorithm is developed to assess the success window of implementing the FLEX strategy, given the uncertainties related to the operator actions (specifically opening the ADVs and aligning FLEX pumps) as well as the underlying phenomena. The AI algorithm is constructed via an artificial neural network (ANN) which is trained using a database of the plant thermal hydraulic response generated by means of the best estimate plus uncertainty (BEPU) methodology. Once trained, the AI meta-model is used to identify the success window of the FLEX strategy under various conditions. The developed meta-model could predict the successful implementation with reasonable accuracy. However, it was not as successful in predicting the failed cases. Nonetheless, AI offers a cost effective computational tool for uncertainty quantification and verification of emergency procedures compared to traditional approaches.