EMD动态过程神经网络尾水管故障信息融合诊断  

Process neural network based on EMD for fault fusion diagnosis of draft tube

作　　者：王瀚[1,2] 张欣伟[2] 罗兴锜[2] 许明海 

机构地区：[1]中国水电顾问集团西北勘测设计研究院,西安710065 [2]西安理工大学动力工程系,西安710048 [3]甘肃酒泉供电公司,甘肃酒泉735000

出　　处：《水力发电学报》2012年第6期282-287,共6页Journal of Hydroelectric Engineering

基　　金：国家自然科学基金项目(50779056;50909081);陕西省普通高等学校重点学科建设项目;陕西省"13115"科技创新工程技术研究中心项目(2008ZDGC-16)资助

摘　　要：为准确识别水轮机尾水管涡带状态,本文提出一种基于EMD指标能量的神经网络故障诊断方法。采用经验模态分解方法分离尾水管压力脉动信号,建立基于指标能量的故障特征向量,并以此作为故障样本训练三层离散前馈过程神经网络,形成由动态特征向量到故障类型的映射关系,实现故障信息融合诊断。以水轮机尾水管压力脉动信号为例,进行了应用检验。结果表明,该方法的收敛速度和分类精度优于传统的RBF、BP神经网络法,分类识别的平均百分比误差小,准确度高,适合用于水轮机尾水管涡带的故障融合诊断。To diagnose accurately vortex rope in the draft tube of hydraulic turbine,this paper presents a new method of fault diagnosis based on empirical mode decomposition（EMD）,index energy and process neural network（PNN）.This method adopts an EMD method to decompose the monitored pressure pulsation signals of draft tube and constructs index energy vectors of the signals.Then it takes those vectors as fault samples to train a three-layer discrete feedforward process neural network,establishes a mapping of dynamic feature vectors into fault type,and finally realizes intelligent fault diagnosis.Application to a practical example shows that the method converges faster and its forecast accuracy is higher than that of traditional RBF or BP neural network.The method produces a small average classification error and hence it is suitable for vortex fusion fault diagnosis of draft tube.

关 键 词：水轮机 尾水管 涡带 EMD指标能量 过程神经网络 融合诊断 

分 类 号：TK730.314[交通运输工程—轮机工程]