Adaptive Multi-Step Compensation-Based Robust Model Predictive Control for Diesel Hydrotreating Reactors With Uncertain Nonlinear Dynamics

Abstract

An adaptive multi-step compensation-based robust model predictive control (RMPC) method is proposed to address the issues of poor control precision of outlet temperature and large fluctuations in cold hydrogen flow in diesel hydrotreating reactors. First, the system model is described as a gray-box model composed of a linear component and an uncertain nonlinear dynamics component. For the linear component, the robust model predictive control method is adopted, into which dynamic correction and rolling feedback mechanisms are integrated, to ensure that the system operates stably under the worst-case conditions. In the latter, a dual-channel feedforward compensation method for uncertain nonlinear dynamics is employed. On the premise of not affecting system stability, state information is fully utilized to compensate for the impacts of factors such as description errors and external disturbances, thereby improving the control accuracy of outlet temperature, predicting changes in control inputs at future moments, and reducing input fluctuations of cold hydrogen flow. Stability and convergence analyses demonstrate that the proposed method achieves stable operational performance. The results of MATLAB simulations and semi-physical platform experiments demonstrate that, compared with the RMPC method and the RMPC with one-step optimal compensation method, the proposed method reduces the MSE by 66.58% and 42.76%, and the IAE by 40.29% and 23.20%, respectively. Note to Practitioners—This paper presents a practical control scheme for industrial diesel hydrotreating reactors aimed at stabilizing the outlet temperature, which is often subject to time-varying delays and uncertain nonlinear dynamics. The proposed adaptive multi-step compensation-based robust model predictive control method is designed with a focus on engineering implementation: it employs a low-order gray-box model that is easier to develop than high-fidelity first-principles models, thereby reducing upfront engineering effort and long-term maintenance costs. This approach enables plant engineers to maintain precise temperature control without frequent retuning or extensive model updates, thus ensuring consistent product quality and extending catalyst service life. Additionally, by actively smoothing the manipulated variable—the cold hydrogen flow rate, the method effectively mitigates excessive actuator movement, thereby reducing mechanical wear, energy consumption, and the risk of thermal fatigue in downstream equipment. Practitioners should note that the initial robust constraints can be tuned using historical plant data, and the strategy is intended to address performance degradation under normal operational variability; it is not a replacement for traditional safety instrumented systems designed for hardware failure protection. The method has been validated on a semi-physical platform simulating real industrial conditions, confirming its readiness for deployment in actual hydrotreating units.