摘要
针对间歇过程的多工况和非线性特征,提出一种基于近邻特征标准化(Nearst Neighborhood Feature Standardization,NNFS)样本的核特征量(Kernel Feature Statistics,KFS)故障检测方法。首先,将间歇过程数据按批次方向展开构成二维建模样本,计算每个样本的局部近邻,采用近邻特征实现标准化,提取多工况批次之间的正常偏差,克服Z-score标准化将多工况过程数据看作一个整体而造成的不准确问题。其次,通过核方法将经过标准化后的样本映射到高维空间,在核空间建立监视模型,计算特征量,并提出采用方差分析(variance,VAR)方法确定核参数,通过核密度估计法确定统计控制限。最后,在青霉素发酵过程进行仿真研究,通过比较表明了所提方法的有效性。
For the multimode and nonlinear characteristics of batch processes, a fault detection method based on kernel feature statistics of samples standardized with nearest neighborhood is proposed. Firstly, batch process is unfolded along batch direction and modeling samples with two dimensions are constructed, local neighborhoods are computed for every sample, standardization is realized with neighborhood feature （mean, standard deviation） and normal errors between batches are got, inaccurate problem caused by Z-score standardization in which multimode data is looked as a mode was overcame. Secondly, samples standardized are mapped high space by kernel method. In kernel space, monitoring model is built and statistics are computed. Furthermore, kernel parameters are determined by variance （VAR） method and control limits of statistics was settled down by Kernel density estimation method. The simulation is realized on penicillin fermentation process and the comparing results show the effectiveness of the proposed method.
出处
《计算机与应用化学》
CAS
CSCD
北大核心
2014年第10期1157-1161,共5页
Computers and Applied Chemistry
基金
国家自然科学基金重点资助项目(61034006)
国家自然科学基金资助项目(61174119
60774070)
辽宁省教育厅科学研究项目(L2012139
L2013155)
辽宁省博士启动基金项目(20131089)
关键词
近邻特征
核主元分析
多工况间歇过程
故障检测
非线性
nearest neighborhood feature （NNF）
kernel principal component analysis （KPCA）
multimode batch process
fault detection
nonlinear
