Iterative Learning Predictive Control Method Based on Reinforcement Learning for Rubber Mixing Batch Process

As a typical batch production process, the rubber mixing process involves complex dynamics and significant fluctuations in results. Since the mixing temperature is key to product quality, this paper proposes an iterative learning model predictive control (ILMPC) strategy that integrates reinforcement learning (RL) for the temperature control model, utilizing the principle of reaction heat balance. The Iterative Learning Control (ILC) enhances performance across repetitive batches. Simultaneously, Model Predictive Control (MPC), leveraging its predictive model, compensates for external disturbances, ensuring primary tracking performance. To address the model accuracy dependence of ILMPC, RL is integrated to enhance the system’s robustness against non-repetitive disturbances and uncertainties through data-driven adaptive learning, optimizing overall performance. Finally, simulation experiments on a rubber mixing temperature control process demonstrate the superior control performance of the proposed method. Note to Practitioners—Precise temperature control in rubber mixing processes remains a significant challenge due to inherent nonlinearity, time delays, and disturbances. This paper presents a method aimed at improving control in repetitive industrial processes by addressing these issues through an adaptive learning mechanism. By leveraging data from previous batches, the proposed approach enhances performance and consistency while effectively handling unexpected disturbances. However, practical implementation may face challenges such as computational complexity, sensor limitations, and variations in material properties. Future research could focus on optimizing the algorithm for real-time applications, improving robustness to external uncertainties, and extending the approach to other batch-based manufacturing processes requiring precise thermal regulation.