摘要
面向过程控制图的模式识别,提出了一个广义神经网络系统.该系统基于广义过程对象模型发生数据,离线训练后能够在线识别各类工业过程常见的控制图模式,模块化的设计使得神经网络系统的结构相对简单,有效地提高了网络的训练速度和模式识别的准确率.首先研究了广义过程对象模型参数对神经网络控制图模式识别率的影响,并基于此影响规律设计了包含模式识别分类模块与模式参数估计模块的集成化神经网络系统结构;其次使用基于广义对象模型产生的数据对神经网络系统进行了训练和验证,讨论了学习训练方法,并进行了控制图模式识别性能的仿真测试,获得了满意的结果.在TE过程仿真平台上进行了实验,给出了对上升阶跃模式和下降阶跃模式的识别结果,表明了具有较高的识别率.
To recognize control chart patterns of processes,a generalized neural network system is proposed.The system is trained on off-line data based on a generalized process model,and it can be used to online recognize common control chart patterns of a variety of industrial processes.In addition,the modular design of the network can lead to simplicity in structure,as well as improvement in performance.First,the influences on control chart pattern recognition performance with different generalized process model parameters are investigated.In this context,an integrated modular neural network system that includes pattern classification modules and pattern parameter estimation modules is developed.Then,the good performance of this system is illustrated with training and testing approaches using data of generalized process model.It is demonstrated on a TE process simulation platform,where recognitions of upshift and downshift patterns are performed with satisfactory results.
出处
《东南大学学报（自然科学版）》
EI
CAS
CSCD
北大核心
2009年第S1期48-52,共5页
Journal of Southeast University：Natural Science Edition
基金
北京市重点学科基金资助项目(XK100100435)
关键词
控制图模式识别
神经网络
广义过程对象
模块化
control chart pattern recognition
neural network
generalized process model
modular structure
