Novel Formulation of Adaptive MPC as EKF Using ANN Model: Multiproduct Semibatch Polymerization Reactor Case Study

Area of Science:

• Chemical Engineering
• Control Systems Engineering
• Artificial Intelligence

Background:

• Nonlinear Model Predictive Control (MPC) is crucial for complex industrial processes.
• Data-driven approaches using Artificial Neural Networks (ANNs) offer adaptive control solutions.
• Integrating Extended Kalman Filters (EKF) with ANNs can improve parameter estimation.

Purpose of the Study:

• To propose a novel nonlinear MPC formulation using an ANN-EKF approach for supervisory control.
• To evaluate the performance of the proposed ANN-EKFMPC strategy in a challenging industrial case study.
• To demonstrate the versatility and effectiveness of a purely data-driven control scheme.

Main Methods:

• Developed a two-module scheme for online parameter and control estimation using ANN-EKF.
• Implemented the ANN-EKFMPC strategy in a cascade configuration with a Proportional-Integral (PI) controller (ANN-EKFMPC-PI).
• Incorporated MPC features like move suppression and terminal constraints for robust control.

Main Results:

• The ANN-EKFMPC-PI controller demonstrated superior temperature tracking and smoother input profiles compared to standard MPC.
• The proposed scheme exhibited enhanced constraint-handling capabilities in simulations for different products and seasons.
• Nominal stability and offset-free tracking were theoretically proven for the controller.

Conclusions:

• The data-driven ANN-EKFMPC-PI strategy provides a versatile and effective solution for nonlinear process control.
• This approach offers significant advantages over traditional MPC methods, particularly in complex industrial applications.
• The integration of ANNs and EKF within an MPC framework enables robust and adaptive control based solely on measurements.