International Journal of Chemical Separation Technology
Dynamic Matrix Control Versus Nonlinear ModelPredictive Control for the Van de Vusse ContinuousStirred-Tank Reactor: Diagnosing and Correcting aCost-Function Pathology That Masked NMPC’sTrue Performance
Alumnus, Carnegie Mellon University, Pittsburgh, Pennsylvania, USA
Abstract
The Van de Vusse reaction in a continuous stirred-tank reactor (CSTR) is a benchmark for nonlinear multivariable control, but its narrow operating envelope creates a trap for nonlinear model predictive control (NMPC): when the setpoint step is small relative to the plant’s reachable range, a standard quadratic NMPC cost function can make zero control action mathematically optimal, even with a perfect prediction model. This paper documents that failure mode, traces it to three compounding causes, and shows correcting them reverses the outcome between Dynamic Matrix Control (DMC) and NMPC. A four-controller comparison – Linear DMC, a neural-network hybrid DMC (NN-DMC), Fixed-Weight NMPC, and an Actor–Critic reinforcement-learning NMPC (A2CRL-NMPC) – initially found both NMPC variants never moved from steady state. Diagnosis traced this to a setpoint too small for the reachable range, a weight ratio letting the move penalty dominate the tracking benefit, an error normalisation that shrank the signal further, and an optimiser trapped near zero effort. Three corrections were applied: the largest reachable setpoint, a rebalanced weight ratio, and a terminal cost normalised by step size, with the optimiser switched to Powell’s method. After these corrections, both NMPC variants settled within 3–5 minutes and reduced normalised squared error 5.4–5.8-fold and absolute error 5.8–6.4-fold versus DMC, while DMC kept a narrow advantage in concentration offset. The RL agent, given six short pre-training episodes, matched but did not exceed the fixed-weight controller. These results give a corrected DMC–NMPC comparison and a diagnostic checklist for NMPC controllers that appear inexplicably inactive.
Keywords: Dynamic matrix control, nonlinear model predictive control, Van de Vusse reaction, CSTR, cost function design, reinforcement learning, actor–critic, controller diagnostics, chemical process control
References (21)
- Shyamalagowri M, Rajeswari R. Modeling and simulation of non linear process control reactor—Continuous stirred tank reactor. Int J Adv Eng Technol. 2013;6(4):1813–8.
- Palanki S, Kolavennu S. Simulation of control of a CSTR process. Int J Eng Educ. 2003;19(3):398–402.
- Qin SJ, Badgwell TA. A survey of industrial model predictive control technology. Control Eng Pract. 2003;11(7):733–64.
- Cutler CR, Ramaker BL. Dynamic matrix control—A computer control algorithm. In: Proceedings of the Joint Automatic Control Conference. San Francisco (CA); 1980. Vol. 17. p. 72.
- Garcia CE, Prett DM, Morari M. Model predictive control: theory and practice—A survey. Automatica. 1989;25(3):335–48.
- Richalet J, Rault A, Testud JL, Papon J. Model predictive heuristic control: Applications to industrial processes. Automatica. 1978;14(5):413–28.
- Mayne DQ, Rawlings JB, Rao CV, Scokaert POM. Constrained model predictive control: Stability and optimality. Automatica. 2000;36(6):789–814.
- Peregrino CA, Lopez Lopez G, Ramirez-Corona N, Alvarado VM, Alvarado Lopez FA, Borunda M. Unified algebraic framework for centralized and decentralized MIMO RST control for strongly coupled processes. Mathematics. 2026;14(4):677.
- Selvamurugan A, Vinod Kumar A, Palan HR, Simiyon A, Indiran T. Reinforcement learning-based nonlinear model predictive controller for a jacketed reactor: A machine learning concept validation using Jetson Orin. ACS Omega. 2025;10(28):30864–78.
- Powell MJD. An efficient method for finding the minimum of a function of several variables without calculating derivatives. Comput J. 1964;7(2):155–62.
- Nelder JA, Mead R. A simplex method for function minimization. Comput J. 1965;7(4):308–13.
- Bhatt DS, Sharma SN. Estimation of the van de Vusse reactor via Carleman embedding. arXiv preprint arXiv:1909.13042. 2019 Sep 28.
- Mayfrank D, Velioglu M, Mitsos A, Dahmen M. Sample-efficient reinforcement learning of Koopman eNMPC. Computers & chemical engineering. 2025 Oct 1;201:109240.
- Aydin E, Bonvin D, Sundmacher K. NMPC using Pontryagin’s Minimum Principle-Application to a two-phase semi-batch hydroformylation reactor under uncertainty. Computers & Chemical Engineering. 2018 Jan 4;108:47-56.
- Aggarwal S. The Intelligent Frontier: Orchestrating AI and Machine Learning for Advanced Financial Fraud Detection. NOLEGEIN Journal of Leadership and Strategic Management. 2026;9(2):30-7p.
- Shipra Aggarwal, Manish Aggarwal. Optimizing Enterprise Financial Risk Management: An Advanced Ensemble-Based Machine Learning Framework for Transactional Fraud Mitigation and Business Continuity. NOLEGEIN Journal of Business Risk Management. 2026; 9(2): 55–67p.
- Aggarwal S. The Intelligent Frontier: Orchestrating AI and Machine Learning for Advanced Financial Fraud Detection. NOLEGEIN Journal of Leadership and Strategic Management. 2026;9(2):30-7p.
- Aggarwal, S., & Aggarwal, M. (2026). What Drives Country-Level Exposure to the Social Cost of Carbon? Discounting, Economic Scale, and Equity in Developing Economies. International Journal of Economics and Financial Issues, 16(5), 26–33. https://doi.org/10.32479/ijefi.24465
- Beal LD, Hill DC, Martin RA, Hedengren JD. Gekko optimization suite. Processes. 2018 Jul 31;6(8):106.
- Nagy ZK, Braatz RD. Robust nonlinear model predictive control of batch processes. AIChE Journal. 2003 Jul;49(7):1776-86.
- Ridlehoover GA, Seagrave RC. Optimization of Van de Vusse reaction kinetics using semibatch reactor operation. Industrial & Engineering Chemistry Fundamentals. 1973 Nov;12(4):444-7.