RESEARCH ON INTERPRETABILITY TECHNIQUES IN TWO WIDELY USED NEURAL NETWORK FOR NUCLEAR POWER PLANT FAULT DIAGNOSIS

In the domain of nuclear power plant operations, effective and reliable fault diagnosis is crucial for ensuring safety and operation efficiency. However, the black-box properties of machine learning models limit their practical application in the field of nuclear power fault diagnosis. This study uses the Shapley value theory in game theory to analyze the reasoning process of the fault diagnosis model, and combines the attention mechanism to propose an innovative and advanced nuclear power fault diagnosis model whose reasoning process is interpretable. A series of experiments based on simulation data were conducted and compared with baseline methods, proving that the method proposed in this study can effectively improve the performance of the nuclear power fault diagnosis model. This research can provide an important reference for the application of machine learning methods in nuclear power fault diagnosis.