Quadratic-Criterion-Based Model Predictive Iterative Learning Control for Batch Processes Using Just-in-Time-Learning Method

• Published in: IEEE access 2019, Vol.7, p.113335-113344
• Main Authors: Zhou, Liuming, Jia, Li, Wang, Yu-Long
• Format: Article
• Language: English
• Subjects: Algorithms; Batch processes; Batch processing; Batch production systems; Computational geometry; Control theory; Convergence; Convexity; Criteria; Data models; Iterative learning control; Iterative methods; just-in-time-learning; local models; Machine learning; model predictive control; Optimization; Predictive control; Predictive models; Real time; Real-time systems; Teaching methods; Tracking control; Tracking problem; Trajectory
• ISSN: 2169-3536; 2169-3536
• DOI: 10.1109/ACCESS.2019.2934474

In this paper, a new quadratic-criterion-based model predictive iterative learning control (QMPILC) algorithm for tracking problem of batch processes is proposed. In the proposed QMPILC design, a parametric time-varying model consisting of a set of local models is established for nonlinear batch processes by using the just-in-time-learning method. In order to describe the processes more accurately, the model is updated with batch running. On basis of the identification model, iterative learning control is combined with model predictive control based on a quadratic performance criterion, and the control law can be obtained by solving a convex optimization problem. According to the real-time feedback information, the input is updated to reject real-time disturbance. As a result, the proposed QMPILC algorithm improves control performance and optimization efficiency. In addition, the convergence and tracking performance of QMPILC are analyzed. The proposed methods are illustrated on batch reactor. The results are provided to show excellent performance of tracking product qualities.