针对单一阈值不能适用于多变工况条件的问题,采用动态阈值修正的流量平衡法与神经网络法相结合的新方法检测反应釜内冷却水盘管泄漏。通过盘管泄漏模拟试验,研究工况变化与泄漏时盘管进出口流量的变化情况。从流量信号中提取盘管泄漏的特征指标构造出神经网络输入矩阵,通过使用大量试验数据对BP神经网络进行训练,对比不同结构的网络训练误差结果,确定其网络结构,建立对盘管运行状况进行分类的BP神经网络模型。试验证明,这种方法能有效检测出盘管泄漏。 Aiming at the problem that a single threshold can not be applied to variable working conditions, a new method combining flow balance method with dynamic threshold correction and neural network method is used to detect the cooling water coil leakage in the reactor. Through the coil leakage simulation test, the changes of the working conditions and the changes of the coil inlet and outlet flow rate during the leakage were studied. The characteristic indexes of coil leakage were extracted from the flow signal to construct the input matrix of neural network. The BP neural network was trained by using a large amount of experimental data. The network training error results of different structures were compared to determine the network structure, The classification of BP neural network model. Experiments show that this method can effectively detect the coil leakage.