For nonstationary processes, it is difficult to detect the abnormality which may be hidden by the normal nonstationary variations. The key issue is how to fully explore the underlying nonstationary variable relationships. In this paper, a sparse cointegration analysis (SCA)-based total variable decomposition and distributed modeling algorithm is proposed for nonstationary processes. First, different cointegration relationships among nonstationary variables are fully decomposed by variable selection with replacement and without replacement based on the SCA. In each block, the nonstationary variables are closely linearly correlated and can be well explored by a local cointegration vector. Different blocks may have some variables in common or have no overlapping variables. Second, the lower level distributed monitoring strategy is proposed to reflect the local nonstationary behaviors of the variables in each block. The upper level monitoring model is constructed to consider the relationships among different nonstationary blocks which have not been addressed by the lower level monitoring models. The feasibility and performance of the proposed method are illustrated with a real industrial process.
Read full abstract