Combinatorial Techniques for Fault Diagnosis in Nuclear Power Plants Based on Bayesian Neural Network and Simplified Bayesian Network-Artificial Neural Network

Knowledge-driven and data-driven methods are the two representative categories of intelligent technologies used in fault diagnosis nuclear power plants. have advantages interpretability robustness, while better performance ease modeling inference efficiency. Given complementarity methods, a combination them is worthwhile investigation. In this work, we introduce new techniques based on Bayesian theory (knowledge-driven) artificial neural network (data-driven) for The first approach exploits an integrated technique, Neural Network (BNN), which introduces into to provide confidence diagnosis. second approach, denoted as Simplified Network-Artificial (SBN-ANN), adopts hierarchical idea, firstly uses simplified diagnose types then severity faults. implemented verified with simulated faults data typical pressurized water reactor. Compared single-algorithmic diagnostic approaches such network, combinatorial show precision. results suggest feasibility develop knowledge dual-drive