Pattern extraction and predictive control based on dynamic controlled autoencoder

doi:10.1016/j.jfranklin.2026.108600

Pattern extraction and predictive control based on dynamic controlled autoencoder

2026-05-01

0

PRE

AI

DOI：10.1016/j.jfranklin.2026.108600

Original

Original request for help

Save

Abstract

En 中文

For modeling and controlling nonlinear industrial processes with high-dimensional process variables, existing latent variable techniques can effectively deal with the high-dimensional nature of the process variables, but the nonlinear dynamics between the pattern and the control inputs are unmodeled, thus limiting modeling accuracy and direct control of the pattern. This paper explores dynamic controlled pattern extraction and a pattern-based control scheme for nonlinear industrial processes. Firstly, a dynamic controlled autoencoder model is proposed to simultaneously perform downscaling of nonlinear process variables and dynamical modeling, enabling the comprehensive capture of the process’s operational information. The introduction of a linear parameter-varying model in the pattern space simplifies the design of the pattern-based controller. Then, a pattern-based model predictive control scheme is developed to enable direct control of the pattern. Finally, the effectiveness of the proposed model for pattern modeling and the pattern-based control scheme is demonstrated through the Tennessee Eastman process and numerical simulation.

Keywords:

dynamic controlled autoencoder

pattern extraction

nonlinear industrial processes

model predictive control

linear parameter-varying model

Pattern extraction and predictive control based on dynamic controlled autoencoder

Abstract

Journal

Organization