基于BP神经网络的盘管泄漏检测方法研究  

Study of the coil-leak detective method based on the BP neural network

作　　者：袁寅[1] 袁昌明[1] 王强[1] 

机构地区：[1]中国计量学院质量与安全工程学院,杭州310018

出　　处：《安全与环境学报》2011年第4期189-192,共4页Journal of Safety and Environment

基　　金：国家自然科学基金项目(60902095);浙江省自然科学基金项目(Y1090672)

摘　　要：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。The present paper is devoted to the study of the coil-leak detective method based on the BP neural network in hoping to extract its boundless application prospect. As a matter of fact, with the ever- increasing chemical safety demands, traditional ottline pipe leak de- tection methods, such as pressure-keeping methods, which fail to meet the needs of on-line detection and control of the leakage of water coil of the reaction kettle for their poor real-time up-to-date perfor- mance. Flow balance method, though still effective in online leak-de- tection, also fails to meet the challenges of the fast-changing working conditions. Therefore, scientists began to face the challenge by using the flow balance method combined with neural network. In order to study the validity of this detection method, we have established an experimental platform of coil leak detection based on $7 - 300PLC. The piatfdrm can not only be able to simulate the leak of water coil, collect the flow data, but also produce warning alarms and help to control some sudden, unexpected leakage. Therefore, we have made an analysis of the flow changes in the inlet and outlet of the water coil of the reactor by means of a series of simulated experiments with the coil leakage, including the fast changing situations of working condi- tions and the leakage variations, we have also extracted characteristic signals （RMS） from the flow signal to protract RMS curve of flow. Careful comparison of the RMS curves of normal, leak and fast changing situations of working conditions, has offered us possibilities to make clear the features of some quite different conditions. We have extracted RMS of the flow to construct the input matrix of the neural network. Through searching for a large number of experimental data to train the BP neural network, it becomes possible to work out the optimal neural network structures by comparing the network training error results of various structures. It is the BP neural network model we have proposed that can help us to dist

关 键 词：安全工程 盘管泄漏 神经网络 流量平衡 泄漏检测 

分 类 号：TP277[自动化与计算机技术—检测技术与自动化装置]