基于PCA-RBF神经网络的小型压水堆故障诊断方法研究

Fault Diagnosis of A Small Reactor Based on PCA-RBF Neural Network

小型反应堆具有结构紧凑和多用途的特点,受到了国际社会越来越多的关注。传统的基于信号阈值的故障诊断方法越来越无法满足精确性和高效性的要求。本文利用RELAP5软件模拟了小型压水堆不同功率水平下的反应堆稳态,蒸汽发生器传热管破裂事故(SGTR)以及冷却剂丧失事故(LOCA),生成了故障诊断所需的样本集。建立了基于PCA-RBF神经网络的故障诊断模型,对不同故障类型、位置和故障程度进行了准确的诊断。最终通过和BP神经网络对比,验证了本文所建立的基于PCA-RBF神经网络的故障诊断模型具有更快的训练速度和诊断精度。

The small modular reactor has the characteristics of compact structure and wide application,which has received more and more attention from the international community.Traditional fault diagnosis method based on signal thresholds has lower accuracy and efficiency.In this paper,the RELAP5 software was used to simulate the steady state,steam generator tube rupture(SGTR)accident and loss of coolant accident(LOCA)under different loads.The sample set required for fault diagnosis was generated.A fault diagnosis model based on PCA-RBF neural network was established.Accuracy of diagnosis of different types of fault,locations and degrees reaches 90%.Compared with BP neural network,RBF neural network has faster training speed and diagnostic accuracy.The established fault diagnosis system was finally proved to have excellent fault diagnosis capabilities.