(149c) A Machine Learning Assisted Approach to Model Predictive Control with Multi-Objective Optimization and Multi-Criteria Decision Making

Model predictive control (MPC) is an advanced control methodology that is widely used in various engineering fields. In chemical engineering, MPC has been extensively studied and applied for several decades [1]. In recent years, the integration of machine learning (ML) with MPC has garnered attention in both academic research and industrial applications. One reason for this increased interest is the difficulty and time required to construct accurate first-principles models using mathematical equations for many complex chemical processes [2, 3]. Furthermore, first-principles models, especially those involving many differential equations, can be computationally expensive to solve. The combination of MPC with ML has demonstrated the capability to improve the performance of control systems and reduce computational burden [4]. Through an extensive literature review, it has been observed that there have been numerous studies on MPC applications in chemical engineering and a growing trend towards incorporating ML into MPC. However, most of these studies only consider a single objective (e.g., deviation from setpoint) in the MPC process; this approach is typically not optimal in the design and operation of complex chemical processes, where multiple objectives (e.g., those related to economic benefits, environmental impact, process safety, energy efficiency, and product quality) need to be considered and optimized simultaneously [5, 6].

In this work, we address the research gap by proposing a comprehensive ML aided MPC with multi-objective optimization (MOO) and multi-criteria decision making (MCDM) methodology (abbreviated as ML aided MPC-MOO-MCDM) in chemical engineering. The proposed methodology is evaluated on a continuous stirred tank reactor (CSTR), considering up to three objectives within the MPC process. The results demonstrate its capability to achieve intended optimization considering multiple objectives in MPC without compromising the convergence of the controlled system. The present work also reinforces the viability of using ML models as surrogates for first-principles models in process control and optimization. Overall, this work exhibits the effectiveness of the proposed ML aided MPC-MOO-MCDM methodology and its applicability to complex chemical processes.

[1] S. Vazquez et al., "Model predictive control: A review of its applications in power electronics," IEEE industrial electronics magazine, vol. 8, no. 1, pp. 16-31, 2014.

[2] Z. Wang, J. Li, G. P. Rangaiah, and Z. Wu, "Machine learning aided multi-objective optimization and multi-criteria decision making: Framework and two applications in chemical engineering," Computers & Chemical Engineering, vol. 165, p. 107945, 2022.

[3] K. McBride and K. Sundmacher, "Overview of surrogate modeling in chemical process engineering," Chemie Ingenieur Technik, vol. 91, no. 3, pp. 228-239, 2019.

[4] F. Simonetti, G. D. Di Girolamo, A. D’Innocenzo, and C. Cecati, "Machine Learning for Model Predictive Control of Cascaded H-Bridge Inverters," in 2022 IEEE 21st Mediterranean Electrotechnical Conference (MELECON), 2022: IEEE, pp. 1241-1246.

[5] G. P. Rangaiah, Multi-objective optimization: techniques and applications in chemical engineering. world scientific, 2016.

[6] M. Park, Z. Wang, L. Li, and X. Wang, "Multi-objective building energy system optimization considering EV infrastructure," Applied Energy, vol. 332, p. 120504, 2023.