Adaptive Iterative Learning Economic Model Predictive Control for Batch Processes With Non-repetitive Disturbances

Iterative learning model predictive control, as an important advanced control method for batch processes, has strong learning capability and closed-loop performance. The traditional iterative learning model predictive control algorithm can effectively eliminate the effect of repetitive disturbances, and at the same time, it is robust to small-scale real-time disturbances. When there is a large real-time disturbance in the controlled system, the economic performance and system stability are usually difficult to guarantee. In this paper, an adaptive iterative learning economic model predictive control strategy for non-repetitive disturbances is proposed to decompose the system dynamics along the iteration direction and time direction, to split the system disturbances into repetitive and non-repetitive parts, and to establish the dynamic economic optimization problems in the batch-to-batch design and within-batch design, respectively. The batch-to-batch design is to apply offline economic optimization based on iterative learning control to eliminate the effects of repetitive disturbances; and the within-batch design is to estimate the non-repetitive disturbances by introducing an extended state observer, and economic model predictive control is implemented online based on the batch-to-batch optimization results, which improves the dynamic economy of the system while suppressing real-time disturbances. The stability of the proposed adaptive iterative learning economic model predictive control strategy is theoretically demonstrated by combining the observer stability analysis method, and the effectiveness of the algorithm implementation is verified by batch reactor simulation experiments.