Research On Model Predictive Iterative Learning Control

Model predictive control (MPC) and iterative learning control (ILC) are both popular algorithms in industrial process control and optimization. The article firstly introduces the generation and development status of MPC and ILC. By listing and analyzing basic algorithms, it describes that ILC cannot deal with nonrepeating disturbances efficiently. Thus, the combination of ILC and MPC is proposed. This paper studies the shortcomings of the existing model predictive iterative learning control (MPILC) and proposes three types of modified MPILC algorithms. The first type of modified algorithms is the fuzzy-model-based nonlinear model predictive iterative learning control (NMPILC). The fuzzy-model-based NMPILC utilizes T-S fuzzy model to give a more accurate description of the nonlinear system. Based on this nonlinear model, the NMPILC focuses on the improvement of classical MPILC in three aspects:kalman filter, predictive horizon and objective function. Then the convergence property of the fuzzy-model-based NMPILC algorithm is presented and through the simulations of a numerical example and a pH neutralization process example, the better control performance of the fuzzy-model-based NMPILC is demonstrated. The second type of modified algorithms is the predictive function iterative learning control (PFILC). Since the existing MPILC has a big amount of calculation and only can be used to slow processes, predictive function control (PFC) with the advantage of structured control law is combined with ILC to deduce online computing time. Finally, by adopting the feedforward-feedback control law, MPC and ILC are appropriately combined to constitute the third type of modified algorithms named the feedforward-feedback MPILC (FF-FB MPILC). The ILC control law acts as a feedforward controller to drive the process close to the desired reference. Meanwhile, the MPC control law acts as a feedback controller to guarantee the system’s robustness to the noises and disturbances. The performance of FF-FB MPILC is illustrated by a steam-boiler system. Simulation results show this algorithm can improve the control effect. In the three types of modified algorithms mentioned above, the fuzzy-model-based NMPILC and PFILC are both combining MPC and ILC to a unified controller to implement control action on the practical object. Meanwhile, the FF-FB MPILC constitutes a feedforward-feedback control law in which ILC acts as a feedforward controller and MPC acts as a feedback controller.