An Accident Diagnosis Method of CFETR Water-Cooled Blanket Based on Deep Neural Network

Abstract

The accident diagnosis of fusion blanket is one of the important issues of fusion reactor safety. In this study, the water-cooled blanket system of China Fusion Engineering Test Reactor (CFETR) is modeled using the RELAP5 code. On the basis of steady-state initialization, several design basis accidents were calculated, including in-vessel loss of coolant accident (LOCA), in-box LOCA, ex-vessel LOCA, and loss of flow accident (LOFA). The RELAP5 calculation results are used as training and validation sets for accident diagnosis. A CFETR water-cooled blanket accident diagnosis method was constructed using a deep neural network based on long short-term memory (LSTM). The 34 blanket parameters simulated by the program within 60 s of the accident occurrence are used as inputs to the model. Diagnostic analysis is conducted on the types, locations, and severity of accidents in the water-cooled blanket. The results indicate that the model can accurately diagnose and obtain detailed information about accidents. Even if a random error of ±10% is added to the input data, the accuracy of the accident classification model is not less than 99.3%, and the errors of the LOCA break size and LOFA pump speed do not exceed 3%. The model has been validated as an effective method for fusion blanket accident diagnosis.