Contents I. Introduction
Model predictive control (MPC) is a powerful control method that is widely utilized in practical industries. MPC, also named receding horizon control, has great advantages in tackling optimization problems for nonlinear industrial processes with input and output constraints online [1], [2], [3]. For the previous research on nonlinear MPC (NMPC) problems, many researchers have adopted the method linearizing nonlinear systems [4], [5], [6]. In this way, the control effect is exchanged at the expense of losing the model accuracy, which is a challenge to the stability of systems. Recently, many scholars have conducted extensive research on learning-based or data-driven MPC problems [7], [8], [9], [10], [11]. It has exerted profound influence on the development of MPC, which is a meaningful work. Ceusters et al. [7] introduced the on-policy and the off-policy multi-objective reinforcement learning (RL) mechanisms that had better control performance compared to the linear MPC. Kheradmandi et al. [8] introduced a closed-loop re-identification approach for MPC, which utilized historical and current data to monitor and update the system model. Hassanpour et al. [9] proposed artificial or recurrent neural networks based dynamic models to tackle the overparameterization problem in the process of MPC implementations. Xu et al. [10] designed a tube-based robust MPC method to tackle optimal control problems of unknown nonlinear dynamics based on the RL framework. Schuurmans et al. [11] introduced a learning-based robust distributional MPC of the constrained Markov jump system with the unknown transition probability. More generally, Dogru et al. [12] proposed that proportional-integral-derivative parameters were tuned through RL with constraints. Li et al. [13] utilized the fuzzy logic method to identify unknown dynamic systems and introduced a fuzzy adaptive inverse optimal scheme. Subsequently, in [14], the inverse optimal output feedback control problem was solved via the state observer based on the fuzzy logic structure.
