Abstract:

Independent component analysis (ICA) method is mainly used to monitor linear and non-Gaussian process, a method named modified independent analysis (MICA) is proposed to improve the non-Gaussian process monitoring performance. The method uses the information of process data to modify the monitoring process monitoring performance. The method uses the information of process data to modify the monitoring index of ICA. Many industrial process variables have characteristics of nonlinear, non-Gaussian and Gaussian mixture distribution. A method which based on LTSA algorithm and the combined index is proposed to solve these problems of the industrial process. Firstly, the local tangent space alignment (LTSA) algorithm is used to achieve the nonlinear dimensionality reduction of sample data. Then MICA and PCA methods are used to obtain non-Gaussian and Gaussian statistics, and the new statistic, which is weighted by these two statistics, is used for process monitoring. Finally, the proposed method has been applied to monitor the Tennessee-Eastman(TE) process and the ethylene cracking furnace to show its efficiency.

Key words: algorithm, principal component analysis, process control, non-Gaussian, modified independent component analysis, local tangent space alignment algorithm, combined index

摘要：

独立成分分析(ICA)方法主要被用来对线性非高斯过程进行监控,为了提高对非高斯过程的监控效果,则利用过程数据信息对ICA的监控指标进行了改进,提出了一种改进的独立成分分析(MICA)方法。许多实际工业过程数据都具有非线性、非高斯与高斯混合分布的特点,为此提出了一种基于LTSA和MICA与PCA联合指标的过程监控的方法。首先采用局部切空间排列(LTSA)算法对样本数据进行非线性降维,然后分别用MICA和PCA方法得到非高斯与高斯统计量,对其进行加权得到新的统计量,并被用于过程监控。最后将该方法应用在田纳西-伊斯曼(TE)过程和乙烯裂解炉的过程监控中,证明了该方法的有效性。

关键词: 算法, 主元分析, 过程控制, 非高斯, 改进的独立成分分析, 局部切空间排列算法, 联合指标