基于时间序列神经网络的蒸汽发生器传热管泄漏程度诊断研究

针对蒸汽发生器U形传热管泄漏，本文提出了一种基于时间序列神经网络对蒸汽发生器传热管泄漏程度进行诊断研究的方法。首先，对核电厂蒸汽发生器U型传热管泄漏进行机理分析，构建其数学模型，提取其泄漏的直接特征参数，再依据Fisher得分法，提取其间接特征参数；其次，通过滑动时间窗口法从预处理后的时间序列数据中生成数据样本，作为时间序列神经网络的输入，并以蒸汽发生器U形传热管泄漏程度信息为标注，基于反向传播（BP）算法对五层神经网络系统进行训练，得到蒸汽发生器U形传热管泄漏的时间序列神经网络模型；最后，模拟核电厂运行过程蒸汽发生器U形传热管泄漏时的时间序列测试数据。仿真结果表明，时间序列神经网络对演变事件的处理具有较好的有效性和较高的泛化能力，对故障程度的诊断研究具有参考价值。 

Diagnosis of Leakage Degree of Steam Generator Tube Based on Time Series Neural Network

Aiming at the leakage of the steam generator U-shaped tube, a method based on time series neural network is proposed to diagnoze the leakage degree of steam generator tube. Firstly, the leakage mechanism of the steam generator U-shaped tube of the nuclear power plant is analyzed, its mathematical model is constructed, the direct characteristic parameters of leakage are extracted, and then the indirect characteristic parameters are extracted according to Fisher score method. Secondly, data samples are generated from the pretreated time series data by sliding time window method, which is used as the input of time series neural network. Based on Back propagation (BP) algorithm, the five-layer neural network system is trained to get the time series neural network model of the leakage of steam generator U-shaped tube. Finally, the time series test data of the leakage of steam generator U-shaped tube during nuclear power operation are simulated. The simulation shows that the time series neural network is with better effectiveness and generalization ability in dealing with evolutionary events, which is of a reference value for fault diagnosis research.