摘要
工业过程数据具有动态、非高斯等特性.独立成分分析(independent component analysis, ICA)既可以分析数据的非高斯形式,又可以极大地去除多变量间的耦合且满足独立性要求.本文引入粒子群算法优化ICA模型参数,自适应地确定独立成分个数.同时,提出一种基于隐马尔科夫链模型(hidden Markov model, HMM)的自适应检测限设计方法,将时间相关数据块的特征信息变化作为过程故障的检测依据.首先利用由时间窗方法确定的独立成分组成监测矩阵来训练HMM模型,旨在提高独立成分间相关性水平的表示能力;然后将得到的HMM模型对监测矩阵进行相关性评估,并在一定容许裕度的基础上设计评估值的自适应因子及检测限,并据此监测特征信息变化,动态地进行在线故障检测.最后, Tennessee Eastman (TE)仿真平台的实验结果表明了所提方法的有效性.
Industrial process data has dynamic,non-Gauss characteristics.The independent component analysis(ICA)not only can analysis the non-gauss data,but also be able to remove the coupling among the multi-variables and meet the independent requirement.The particle swarm optimization(PSO)algorithm is introduced in this paper to optimize the parameters of ICA model,and the independent components are adaptively determined by negative entropy maximization.Meanwhile,a new adaptive detecting limit based on hidden Markov model(HMM)is put forward to monitor the industrial process,which pays attention to the change of feature information in correlated data blocks.In this proposed method,the independent components in a time window are seemed as the basic unit to train the HMM model,which is contributes to describe correlation evaluation information.With the evaluation value,an adaptive factor and detecting limitation which are based on the acceptable margin are designed to track the change of feature information and monitor the process online dynamically.Finally,the experimental results of the Tennessee Eastman(TE)simulation platform show the effectiveness of the proposed method.
作者
王培良
叶晓丰
杨泽宇
WANG Pei-liang;YE Xiao-feng;YANG Ze-yu(School of Engineering,Huzhou University,Huzhou Zhejiang 313000,China;College of Electronic and Information,Hangzhou Dianzi University,Hangzhou Zhejiang 310018,China)
出处
《控制理论与应用》
EI
CAS
CSCD
北大核心
2018年第9期1331-1338,共8页
Control Theory & Applications
基金
国家自然科学基金项目(61573137)资助~~
关键词
独立成分分析
粒子群算法
隐马尔科夫模型
相关性评估
自适应因子
故障检测
independent component analysis
particle swarm optimization algorithm
hidden Markov model
correlation evaluation
adaptive factors
fault detection