Abstract:

A fault detection method for chemical process based on LPP-GNMF algorithm is proposed. NMF(non-negative matrix factorization) is a novel dimensionality reduction algorithm, with characteristics of positive pure additivity of latent variables in the mechanism, thus, when compressing the data, the information can be described based on the local characteristics inner the data. Compared to the traditional multivariate statistical process monitoring methods such as principal component analysis(PCA), NMF offers a better ability for data explanation. However, firstly, NMF requires the original data to meet the requirements of non-negative, which can not be guaranteed in the actual chemical process, in order to relax the non-negative requirements of the original data, a generalized non-negative matrix factorization(GNMF) algorithm is quoted. Secondly, GNMF does not take the local structure and geometric properties into account during the process of decomposition, which may not be accurate to deal with the problem of data. Aiming at this problem, the algorithm of combining GNMF with LPP(locality preserving projection) is proposed. The proposed LPP-GNMF algorithm is applied to the Tennessee Eastman process to evaluate the monitoring performance. The simulation results show the feasibility of the proposed algorithm compared with the PCA algorithm, the NMF algorithm and the SNMF algorithm.

Key words: algorithm, fault detection, principal component analysis, generalized non-negative matrix factorization, locality preserving projection, simulation

摘要：

提出了基于LPP-GNMF算法的化工过程故障监测方法。非负矩阵分解（NMF）是一种新兴的降维算法，由于它在机理上具有潜变量的正向纯加性的特点，所以在对数据进行压缩时，可以基于数据内部的局部特征有效描述数据信息，相比于传统的多元统计过程监控方法如主元分析（PCA）等有更好的解释能力。然而NMF要求原始数据满足非负性的要求，实际的化工过程有时并不能保证，为放宽对原始数据的非负要求，引入了广义非负矩阵分解（GNMF）算法。其次，GNMF在分解的过程中没有考虑到样本间的局部结构和几何性质，可能存在不能准确处理数据的问题。针对这一问题，提出了将GNMF与LPP（局部投影保留）相结合的算法。将提出的LPP-GNMF算法应用于TE过程来评估其监测性能，并与PCA算法、NMF算法、SNMF算法进行比较，仿真模拟结果表明所提算法的可行性。

关键词: 算法, 故障监测, 主元分析, 广义非负矩阵分解, 局部投影保留, 模拟