Linear and nonlinear hierarchical multivariate time delay analytics for dynamic modeling and process monitoring

Abstract

Due to the complexity of industrial processes, the collected data show typical nonlinearity and dynamic characteristics, bringing significant challenges for nonlinear dynamic process monitoring. Due to the instinctive structure of equipment and different positions of measurements, time delays widely exist between variables, decreasing the accuracy of models. In this work, a linear and nonlinear hierarchical modeling method is proposed for time delay analytics and nonlinear dynamic process monitoring. First, the variables are automatically divided into several linear subgroups and nonlinear variables by extracting dynamic latent variables. Then an autoregressive autoencoder (ARAE) model is designed to describe linear and nonlinear characteristics combined with dynamic-inner principal component analysis (DiPCA). In this way, time delay analytics based on the dynamic framework is developed to enhance the effectiveness of extracting dynamic characteristics. Finally, a hierarchical monitoring strategy is developed for nonlinear processes from both linear and nonlinear, static and dynamic perspectives. The effectiveness is verified by a numerical case and a three-phase flow process.