A geometry constrained dictionary learning method for industrial process monitoring

Abstract

Data-driven process monitoring methods have attracted great attention due to the case that it can provide an efficient way to cope with the industrial process without the need of first-principle models. The local information that results from the reaction process often influences the process monitoring result. Unfortunately, this local information is usually ignored in the data-driven process monitoring method. In this paper, a geometry constrained dictionary learning (GCDL) method is proposed to address the above problem. By exploiting the underlying characteristics and simultaneously holding the local geometrical information, the GCDL method leads to a balance between the reconstructive item and the discriminative item. Inspired by the manifold method, the GCDL method can provide discriminative sparse coding, which means that samples that belong to the same class will share similar sparse coding. It is in accordance with the fact that the samples of different categories often reside on different subspaces embedded in a high-dimensional feature space. The effectiveness of the proposed method is evaluated through two numerical simulation cases, a continuous stirred tank reactor (CSTR) case and a real industrial aluminum electrolysis process. Extensive experimental results reveal that the proposed method shows better performance by taking the local geometrical information into consideration.